Enhancing Occupational Safety Through AI: A Review of Key AI Technologies

Sahar Yousefi; Bhaskar Aryal; John Shutske; Salah F. Issa

doi:10.13031/ja.16502

¹Department of Agricultural and Biological Engineering, University of Illinois at Urbana-Champaign, Urbana, Illinois, USA.

²College of Agricultural, Consumer and Environmental Sciences, University of Illinois at Urbana-Champaign, Urbana, Illinois, USA.

³Department of Biological Systems Engineering; University of Wisconsin-Madison, Madison, Wisconsin, USA.

The authors have paid for open access for this article. This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License https://creative commons.org/licenses/by-nc-nd/4.0/

Submitted for review on 29 July 2025 as manuscript number JASH 16502; approved for publication as a Review Article and as part of the LLMs and Generative AI Applications in Agricultural and Biological Engineering Collection by Community Editor Dr. Michael Pate of the Ergonomics, Safety, & Health Community of ASABE on 4 December 2025.

Citation: Yousefi, S., Aryal, B., Shutske, J., & Issa, S. F. (2026). Enhancing occupational safety through AI: A review of key AI technologies. J. ASABE, 69(1), 165-179. https://doi.org/10.13031/ja.16502

ABSTRACT. Artificial intelligence (AI) has emerged as a transformative tool in various industries, including agriculture, where it has the potential to enhance safety and reduce injury risks. This review explores the application of AI techniques in occupational safety with a special focus on agricultural safety, focusing on predictive modeling, large language models (LLMs), computer vision, and wearable technologies. Due to the limited number of studies that address AI in agricultural safety, research from related fields, such as construction safety, was also considered for its potential applicability. The findings indicate that (1) predictive models that leverage machine learning (ML) algorithms can assess historical data and forecast hazards, enabling proactive safety measures. (2) LLMs can improve injury report analysis by extracting key terms and patterns to identify recurring risks. (3) Computer vision and remote sensing technologies enhance environmental monitoring by detecting real-time unsafe conditions. At the same time, (4) AI-powered wearable devices can track worker health indicators such as heart rate, potentially preventing injuries. A total of 85 studies were analyzed, providing insights into the diverse applications of AI in mitigating occupational hazards, specifically agricultural hazards. This review highlights the current advancements and future research opportunities for AI-driven safety interventions in high-risk occupational environments.

Keywords.Agricultural safety, Artificial intelligence, Computer vision, Health, Injury prevention, Large language models, Machine learning, Occupational safety, Predictive modeling, Wearable technology.

Agriculture, the bedrock of human civilization, is vital in sustaining global economies and ensuring food security for billions of people (Gibson, 2012). Yet, despite its indispensable importance, agriculture remains one of the most hazardous industries worldwide (OSHA, 2025; Bureau of Labor Statistics, 2023; McCurdy and Carroll, 2000; Li et al., 2024; Raza et al., 2024). From the early days of subsistence farming to today’s technologically advanced operations, farmers and agricultural workers have faced myriad dangers (Nguyen et al., 2018). While agricultural innovations have vastly improved productivity, they have also introduced new risks, making safety a persistent concern in the field (Nguyen et al., 2018).

It is crucial to emphasize the importance of ensuring agricultural workers' safety, health, and well-being. Agriculture is a physically demanding profession, requiring workers to perform strenuous tasks under diverse and often unpredictable conditions. Injuries related to heavy agricultural equipment and machinery are a significant concern, as the hazards associated with them pose serious risks if not handled properly (Raza et al., 2024). These workers face daily risks of injury and death while performing essential duties to maintain the global food supply (Nguyen et al., 2018). The agricultural industry has suffered from high and stagnant injury rates in the last three decades despite implementing safety protocols and intervention programs (Lehtola et al., 2008; Issa et al., 2019). In the United States alone, agriculture ranks as one of the top three industries for worker fatalities, with many incidents related to tractor rollovers, auger entanglements, grain entrapment, chemical exposure, and livestock handling (OSHA, 2025). Beyond agriculture, occupational safety is fundamentally about safeguarding the well-being of workers in all industries by identifying hazards, implementing preventive measures, and fostering a culture of safety. Effective occupational safety programs reduce workplace injuries, illnesses, and fatalities, while also improving productivity. By assessing work environments and enforcing safety protocol, ranging from proper machine guarding to hazard communication, organizations can minimize both acute incidents and long-term health consequences (Pishgar et al., 2021).

In practice, traditional safety programs are largely compliance-centric, and their effectiveness is limited by their reliance on human agency. Such dependency on human behavior to notice and self-report incidents leads to systematic underreporting in safety records and conceals leading indicators of occupational risk (Probst et al., 2019; Issa et al., 2016) Consequently, the absence of comprehensive data constrains the implementation of proactive safety management. This fundamental limitation can be addressed by integrating artificial intelligence (AI) into safety frameworks. AI does involve an upfront investment and predictable operating costs, but these are often outweighed by the economic burden of incidents. With the evolution of precision agriculture, autonomous vehicles, and operational autonomy, human-machine interactions have expanded in both scope and risks (Lincoln et al., 2025; Shutske et al., 2025a,b). The dynamic nature of this interaction calls for continuous, real-time monitoring to ensure safe field operations, possible only with AI-driven intelligent systems. For instance, ergonomic assessments like NIOSH Lifting Index, RULA/REBA, and ISO 11228 thresholds can now be computed seamlessly using computer vision systems (Agostinelli et al., 2024).

In recent years, artificial intelligence (AI) has emerged as a potentially transformative technology across various industries, offering advanced tools to enhance safety and mitigate risks in multiple industries, including agriculture (Pallathadka et al., 2023). AI has the ability to process vast amounts of data and make predictive insights. This quickly evolving technology holds significant potential to improve safety outcomes in agriculture (Kakhki et al., 2019a). The application of AI in this domain is particularly promising because of its ability to analyze complex multi-modal data from numerous sources, including machinery, weather conditions, and worker behaviors (Rathod et al., 2024). AI can predict when and where incidents are most likely, enabling preventive measures to save lives and reduce injury rates (Rathod et al., 2024).

This review examined the application of artificial intelligence (AI) methods in multiple occupational domains, including construction, transportation, and agriculture, with a particular emphasis on agricultural safety. This study aims to provide a comprehensive analysis of how AI has been applied to enhance safety across different occupational fields such as agriculture and the construction industry. The key AI techniques employed in these studies are discussed, including predictive models, large language models (LLMs), computer vision technologies, and wearable devices. By compiling and analyzing the existing literature, this study aims to shed light on the current state of AI in occupational safety and highlight the opportunities for further development and application of these technologies.

Materials and Methods

This review examines the application of various AI techniques in occupational safety by analyzing existing research on predictive modeling, large language models (LLMs), computer vision, and wearable technologies. A structured approach was used to identify, categorize, and synthesize studies that explored AI's role in hazard prediction, injury prevention, and real-time risk assessment. Relevant literature was located using key terms such as AI + occupational safety, AI + construction safety, AI + workers safety, workers injury + prediction AI, Computer vision +tractors, AI + agricultural safety, AI + agricultural injuries, injury prediction, agricultural injury prediction, VR + ag safety, VR + safety training, wearable devices + ag safety, wearable devices + health, and worker injury prediction, among others. Key search terms were identified through an iterative process that combined domain expertise in occupational and agricultural safety with commonly used AI terminology. The initial search used broad terms (e.g., AI + occupational safety), which were refined based on recurring keywords and concepts in relevant literature, ultimately capturing three main areas: AI techniques, safety domains, and safety-related outcomes. Relevant literature was identified and conducted in major academic databases, including Scopus, IEEE Xplore, and Google Scholar. Due to the limited number of studies specifically focused on the application of AI in agricultural safety, the scope of this review was extended to include research from a comparable occupational domain with a similar hazard taxonomy, construction safety. Both domains involve decentralized outdoor worksites, heavy equipment, power machinery, exposure to weather conditions, and human-machine interactions. Shared incidents such as falls from height, equipment rollover, entanglement, ergonomic strain etc. occur through similar hazard mechanisms, allowing for transfer of safety knowledge. The domain was also selected because the AI methodologies and techniques it employs have strong potential for adaptation to agricultural safety. This transferability is evident in applications like vision-based hazard detection, ergonomic posture assessment, and distraction monitoring. The review included peer-reviewed journal articles, conference papers, and arXiv preprints published primarily after 2019 to capture recent developments in the field. Studies were eligible if they applied artificial intelligence, machine learning, computer vision, or wearable technologies to safety, health, or injury prevention in agricultural or related work environments, and if they reported methodological details or empirical results. Studies unrelated to agriculture or occupational safety (e.g., medical imaging), as well as editorials, opinion pieces, and non-empirical reviews, were excluded. Duplicates, non-English publications, and sources without accessible full text were also removed from consideration. A total of 85 studies were reviewed. Many risk assessment, injury prevention, and worker monitoring strategies used in this field can be applied to mitigate hazards in agricultural environments and enhance safety for agricultural workers. By incorporating insights from these areas, this review provides a broader perspective on how AI-driven solutions can be leveraged to reduce injury risks in agriculture. After reviewing and analyzing a wide range of studies, those that specifically applied AI techniques in agricultural safety or employed methods with potential applicability in this field were included. To ensure diversity and avoid redundancy, studies that used the identical AI techniques were not included multiple times.

Four key areas were investigated in this review: predictive models, LLMs, wearable devices, and computer vision, each contributing to different aspects of agricultural safety through AI-driven solutions. Predictive models can be applied in agricultural safety by utilizing machine learning (ML) algorithms to analyze historical data on accidents, weather conditions, equipment malfunctions, and worker behaviors. By identifying correlations and trends, these models provide early warnings of potential hazards, enabling proactive safety measures (Kakhki et al., 2020a).

LLMs have been investigated for their capacity to enhance agricultural safety by improving the organization, retrieval, and analysis of injury reports. These models process large volumes of unstructured text data, extracting key terms and patterns to identify recurring risks. By automating data classification and retrieval, LLMs facilitate efficient analysis of safety incidents (Muller et al., 2024). Additionally, customizing fine-tuned pre-trained LLMs based on current safety policies and updated training guidelines can support agricultural workers through user-friendly tools such as chatbots.

Wearable AI-powered devices represent another emerging area in agricultural safety. Studies included in this review examine the effectiveness of smart watches, vests, and wristbands embedded with sensors that monitor vital signs, movement patterns, and environmental conditions (Etienne et al., 2024). These wearables can detect early indicators of fatigue or falling from height, triggering alerts, calling emergency contacts, and, in some cases, halting autonomous machinery operations to prevent injuries. The reviewed studies' methodologies and key findings are discussed in subsequent sections, categorized according to AI technique.

Computer vision, including remote sensing technologies, has also been explored in the reviewed studies for its role in monitoring hazardous conditions. Drones, satellite imagery, and ground-based sensors equipped with AI algorithms provide real-time detection of unsafe environments, such as chemical exposure zones, unstable terrain, and machinery malfunctions. These systems generate automated alerts to mitigate risks before accidents occur (Tian et al., 2020). Figure 1 provides a visual summary of the AI-driven technologies analyzed in this review.

Results

A total of 85 articles were included in the final review, representing a broad range of AI techniques: 10 providing overviews, 17 focusing on predictive models, 16 examining LLMs, 22 addressing wearable technologies, and 20 exploring computer vision applications. 78% of the reviewed articles were published between 2019 and 2025, while 22% were published before 2019, which were mostly foundational studies that established the basis for later advancements in AI, machine learning, and agricultural safety research. These findings provide an overview of how AI-driven solutions have been explored and implemented to enhance safety in occupational settings to date, with special emphasis on agriculture. Within predictive models, studies focused on classification, feature importance analysis, and injury prediction. The LLM category encompassed studies on AI-powered farmer assistance and the analysis of injury reports. Computer vision applications were categorized into environmental risk monitoring and hazard identification, while wearable device studies primarily addressed hazard detection and worker monitoring. Below, four areas of AIs are discussed based on their common application in occupational safety. Figure 2 shows four main categories and their subcategories.

Predictive Models in Occupational Safety

ML Predictive modeling has gained significant traction in occupational safety and health, offering sophisticated ways to anticipate and mitigate potential risks (Pishgar et al., 2021). These models, ranging from those that are based on linear regression methods to more complex ensemble methods like random forests and gradient boosting, enable proactive identification of hazardous events by analyzing historical data and identifying patterns indicative of risk (Jain et al., 2023). Different predictive models have been used in occupational safety due to their effectiveness and adaptability, each suited to specific data types and requirements. Logistic regression, a linear model for binary outcomes, calculates the probability of an event occurring and could be applied in agricultural safety due to its interpretability and capacity to handle categorical variables (Murphy, 2012). Random Forest, an ensemble method that constructs multiple decision trees, is highly robust against overfitting and performs well on large datasets with complex feature interactions. Its ability to capture nonlinear relationships and effectively handle categorical variables makes it especially valuable for agricultural safety modeling (Breiman, 2001; Li, 2022). Gradient boosting and XGBoost are iterative ensemble methods that create sequential models to correct previous errors, making them highly accurate for incident prediction in complex scenarios with diverse features (Badarinath et al., 2021). Support Vector Machines (SVM) are well-suited for classification and regression, particularly with high-dimensional data, and excel in distinguishing classes in sparse or imbalanced agricultural safety data (Kakhki et al., 2019b). These models are vital for predicting rare and binary safety-related events (Ogundimu, 2019). The following sections explorer some of the applications of predictive models in occupational safety.

Figure 1. Topics and techniques examined in the review. Articles related to applications of AI in agricultural safety and related fields were categorized into four categories based on the AI tools investigated. Each category represents distinct application areas relevant to agricultural safety.

Figure 2. The four main AI categories with potential applications in agricultural safety, along with their respective subcategories based on specific use cases. The number of reviewed articles for each category is indicated below each category.

Application of Predictive Models in Occupational Safety

Predicting Injuries and Incidents

Predictive modeling has shown significant potential in predicting the severity of injuries and incidents in occupational safety. Researchers have classified incident severity levels and identified critical injury patterns that inform targeted safety interventions by applying ML predictive models. As an example, Kakhki et al. (2020a) applied Naive Bayes and Random Forest models to classify occupational incidents in agro-manufacturing by severity levels, offering a framework for resource prioritization. In a related effort, Kakhki et al. (2019b) used Support Vector Machines (SVM) and Naive Bayes to assess injury severity in agribusiness settings, illustrating the ability of these models to detect patterns in injury data. Another study by Davoudi (Kakhki et al., 2020b) focused on post-incident injury severity prediction using CHAID and CART classifiers, emphasizing the role of feature selection in improving classification outcomes. Additionally, Liu et al. (2021) addressed class imbalance in severity prediction using resampling techniques and five-fold cross-validation, highlighting essential strategies for improving model performance in rare-event scenarios. Taken together, these demonstrate how predictive models, when supported with appropriate data handling and evaluation techniques, can enhance injury severity prediction and guide prevention strategies prioritized around high-risk incidents.

Identify Injury Risk Factors

Predictive modeling has also been widely used to identify key risk factors associated with fatal and non-fatal occupational incidents. For instance, Zhu et al. (2023) analyzed 694 fatal construction incidents reported by the National Institute for Occupational Safety and Health (NIOSH) to predict characteristics such as the nature of injury and source of exposure using models like Random Forest (RF) and Stochastic Gradient Boosting (SGBM), achieving up to 84.9% accuracy. Choi et al. (2020) applied multiple classifiers, including logistic regression, decision tree, and AdaBoost, to predict fatality likelihood in over 137,000 industrial injury incidents, with RF yielding the highest AUROC of 0.9198 after handling data imbalance with oversampling. In the agricultural domain, Kakhki et al. (2019a) utilized neural network models such as Multi-Layer Perceptron (MLP) and radial basis function networks to assess safety risks and prioritize interventions in grain elevators. These studies highlight how predictive models, when paired with feature selection, interpretability tools, and hyperparameter tuning, can uncover high-impact risk factors and guide evidence-based decision-making in occupational safety management.

The specific methodologies and performance metrics of these reviewed articles are synthesized in table 1.

Large Language Models in Occupational Safety

LLMs represent a transformative advancement in AI, enabling computers to process, understand, and generate human language with remarkable accuracy and nuance. Trained on vast text corpora, LLMs excel in a range of natural language tasks, including translation, summarization, question-answering, and advanced linguistic analysis. These models rely on extensive neural network architectures that leverage billions of parameters, allowing them to perform sophisticated language-based reasoning, context comprehension, and content generation (Min et al., 2023). In agricultural safety, LLM-based methods are beginning to make meaningful strides, enhancing efficiency and safety by assisting with complex tasks involving human communication, data synthesis, and knowledge sharing. They have strong potential to process unstructured data, such as incident reports and safety guidelines, and extract informative keywords and patterns from these texts (Shutske, 2023). Popular LLMs have become foundational tools for natural language understanding and generation across diverse domains, including occupational safety. One notable example is LLaMA (Large Language Model Meta AI), developed by Meta. It offers scalable, efficient language processing capabilities with comparatively smaller model sizes, making it well suited to research settings with limited computational resources (Touvron et al., 2023). ChatGPT, built on OpenAI’s GPT architecture, is renowned for its conversational fluency and is widely used in virtual assistant and chatbot applications. In occupational safety, ChatGPT can deliver interactive, context-sensitive guidance to farmers, enhancing decision-making and compliance (Roumeliotis and Tselikas, 2023). Another influential model is BERT (Bidirectional Encoder Representations from Transformers) by Google, which processes language bidirectionally, allowing it to interpret technical and procedural language more accurately, especially useful in understanding occupational safety guidelines or protocols (Jawahar et al., 2019). As LLMs continue to evolve, newer iterations (e.g., GPT-4, LLaMA 3) are demonstrating even more nuanced reasoning, adaptability, and task generalization. The following paragraphs show the potential application of LLMs in occupational safety.

Application of LLMs in Occupational Safety

Incident Reports Analysis

LLMs are emerging as powerful tools for analyzing incident reports and enhancing injury surveillance, with increasing relevance to agricultural safety. Ahmadi et al. (2025) employed state-of-the-art LLMs, including GPT-3.5, GPT-4.0, Gemini Pro, and LLaMA 3.1, to process OSHA construction injury reports from 2002 to 2023. These models automatically classified key attributes such as root cause, severity, and incident timing. Using a zero-shot learning framework and standardized prompts, GPT-4.0 demonstrated the highest performance overall, while Gemini Pro and LLaMA 3.1 showed strengths in injury cause and severity classification, respectively. This study illustrates the ability of LLMs to extract structured information from unstructured text with high accuracy. In a related effort, Muller et al. (2024) explored LLMs for automating the extraction of injury details from agricultural news reports curated by AgInjuryNews. By evaluating models like GPT-3.5, GPT-4, and a fine-tuned LLaMA 2, the researchers demonstrated that fine-tuning enhances contextual understanding and accuracy, with the fine-tuned LLaMA 2 achieving the best results. This approach has the potential to significantly reduce the manual burden of injury data collection and analysis while maintaining quality and consistency. These studies underscore the transformative potential of LLMs in processing unstructured safety data, enabling timely insights, automating injury surveillance, and informing proactive safety measures in high-risk occupational environments.

Safety Education

LLMs are increasingly being explored as tools for enhancing safety education and decision-making in occupational safety contexts. Johnson and Wilson (2024) investigated the application of GPT-4 for intelligent agricultural machinery management, introducing a multi-round prompt methodology to generate refined, context-aware responses. This iterative approach allowed GPT-4 to provide accurate and actionable insights by breaking down complex machinery management tasks into a sequence of logically structured prompts, outperforming single-prompt queries and other LLMs like LLaMA and ChatGPT in precision and relevance. Their findings suggest that such an approach not only improves operational decision-making but also reduces on-site risks, thereby contributing to a safer agricultural infrastructure. Complementing this, Shutske (2023) examined broader applications of generative AI in agricultural safety education. The study identified five key use cases for LLMs: answering safety-related queries, interpreting scientific safety science literature, summarizing incident reports, streamlining regulatory compliance actions, and generating educational content for outreach. While highlighting the transformative potential of LLMs for educators and safety professionals, the study also pointed to critical concerns around model accuracy, bias, and legal considerations. These studies demonstrate that LLMs can serve as valuable tools in advancing occupational safety education and supporting informed, real-time decision-making when used with iterative, context-sensitive prompting and implemented responsibly.

Developing Chatbot

LLMs are increasingly being utilized to develop intelligent chatbots that enhance safety education through interactive learning. Zaidi et al. (2024a) introduced iSafe Chatbot, an AI-powered educational tool designed to support construction safety training by integrating natural language processing and LLMs. The chatbot responds to user queries about Occupational Safety and Health Administration (OSHA) regulations and provides video-based instructional content, offering a more engaging alternative to traditional static materials. It employs a hybrid architecture that combines a structured safety knowledge base with the generative capabilities of an LLM, enabling contextually relevant and accurate responses. User testing demonstrated that the chatbot significantly improved comprehension and engagement. This approach holds strong potential for adaptation in agricultural safety, where interactive, AI-driven tools can help workers and educators access real-time safety information, clarify complex procedures, and reinforce best practices in high-risk environments.

Data Generation and Annotation

Data collection and labeling are critical steps in training machine learning models, yet they are often resource-intensive, costly, and time-consuming. Collecting high-quality datasets requires extensive fieldwork, specialized equipment, or access to proprietary sources, while accurate labeling demands significant human effort and domain expertise. These challenges are particularly pronounced in agricultural safety, where datasets often need to capture diverse environmental conditions, equipment types, and human activities (Northcutt et al., 2021; Snorkel AI, 2024). LLMs, with their advanced generative capabilities, offer innovative solutions to these challenges. By simulating and augmenting datasets, LLMs can help generate diverse, high-quality data that reduces the burden of manual collection. For instance, LLMs can assist in the transformation of data representations, such as converting textual data into structured formats, or generate synthetic scenarios (Nada? et al., 2025). Paired with generative models like Generative Adversarial Networks (GANs) (Goodfellow et al., 2020) or diffusion models (Goel et al., 2024), they enable the generation of additional image training data. Examples include altering temporal or environmental conditions, such as converting a daytime photo of a tractor in the field into a nighttime equivalent or transforming summer field images into winter conditions with snow. This augmentation increases the variability and robustness of datasets, improving the generalizability of downstream ML models. Table 2 summarizes the studies that implemented LLMs in occupational safety.

Wearable Devices for Human Safety in Occupational Settings

Wearable technology has rapidly evolved, becoming a vital tool for enhancing human safety in various sectors, including healthcare, construction, and agriculture. These devices are designed to monitor physiological and environmental parameters, provide real-time feedback, and reduce injury risks through preventive interventions. In agriculture, the implementation of wearable devices addresses the unique challenges posed by physically demanding tasks and hazardous environments.

In agriculture, wearable devices are being increasingly adopted to address the high injury rates associated with tasks like operating machinery, handling livestock, and working in extreme weather conditions. Commercially available devices like smartwatches monitor workers' health metrics or fatigue (Patel et al., 2022). Heat exposure monitors detect signs of heat stress and dehydration. Wearable cameras or smart glasses with onboard vision systems provide real-time feedback on hazards, such as moving machinery or uneven terrain, helping workers avoid them (Wagoner et al., 2020; Shakerian et al., 2021). These devices reduce physical strain, improving both safety and productivity. Data from these sensors is processed in real time using edge computing, where algorithms embedded in the device analyze the data locally. If certain thresholds are exceeded, such as a high heart rate in extreme heat conditions, the device sends alerts to the user, prompting immediate action.

Application of Wearable Devices in Occupational Safety

Heat Stress Monitor

Wearable sensors allow continuous physiological monitoring, enabling early detection of heat stress, particularly in labor-intensive industries such as agriculture and construction, where workers are often exposed to extreme environmental conditions. These devices not only offer real-time monitoring but also provide critical data for predictive modeling and safety intervention design. One study conducted in Northern Mexico during the grape harvest season utilized ingestible core body temperature sensors and wearable monitors to assess heat stress and hydration among agricultural workers (Wagoner et al., 2020). Workers were provided with CorTemp Ingestible Core Body Temperature Sensors (HQ Inc., Palmetto, FL, USA) and wore waist-mounted monitors that recorded core body temperature every 10 seconds throughout their workday. This approach allowed investigators to track heat exposure and hydration levels across different seasons, revealing that the majority of workers were dehydrated post-shift (urine S.G. >1.020). The study highlighted the vulnerability of migratory agricultural workers to heat-related hazards and underscored the importance of monitoring tools in identifying at-risk populations and designing targeted interventions. Another study explored the use of a wristband-type biosensor to monitor heat strain among construction workers (Shakerian et al., 2021). Physiological signals collected from 18 subjects performing tasks under three distinct environmental conditions were processed using advanced signal decontamination techniques and resampled into arrays of informative features. Supervised machine learning algorithms were then employed to classify individuals’ heat strain states, achieving prediction accuracies exceeding 92%. The study demonstrated the potential of wearable biosensors to continuously assess heat strain and highlighted their applicability in reducing heat-related illnesses and fatalities in construction settings. These findings are highly relevant to the agricultural sector, where similar predictive models could be utilized to prevent dehydration and heat stress among workers. For example, if a worker exhibits signs of fatigue combined with elevated environmental temperatures, the device might recommend hydration breaks or reduced physical activity. Additionally, such data-driven insights could inform the development of tailored safety training sessions, enhancing worker resilience to extreme environmental conditions.

Enhancing Safety and Reducing Work-Related Risks with Exoskeletons

Exoskeletons are wearable devices designed to augment human physical capabilities by providing external support and reducing physical strain. Their primary purpose is to assist the human workforce while mitigating the risk of Work-Related Musculoskeletal Disorders (WMSDs), which are prevalent in physically demanding occupations like agriculture. Although exoskeletons have significant potential to improve safety and productivity, their adoption in agriculture is still limited because of the challenging nature of farm tasks and the demanding environmental conditions in which they are carried out (Arachchige et al., 2024).

Types of Exoskeletons and the Role of AI

There are advantages and disadvantages for farmers to adapt each type of exoskeleton (table 3).Exoskeletons can be broadly categorized into three types based on their power systems:

Active Exoskeletons: These are powered devices equipped with motion sensors, actuators, and advanced control systems. Active exoskeletons provide real-time assistance with a more "natural sensation" of movement (Arachchige et al., 2024). AI plays a significant role in these systems by:

Wristbands for Tracking Health Metrics

Smartwatches and other wristbands have proven useful in healthcare by tracking health metrics, monitoring physical activity, and detecting emergencies (Lu et al., 2020; Reeder and David, 2016; Sengül et al., 2022). In agriculture, these technologies show promise for enhancing safety, particularly by detecting incidents and notifying emergency contacts or emergency medical services (EMS). This capability is especially crucial in isolated rural areas. A study evaluated the Apple Watch Series 7 and Garmin Vivoactive 4 across five simulated agricultural incident experiments, including UTV ejections, forklift drops, skid steer impacts, tractor rollovers, and mower upsets (Etienne et al., 2024). The devices used accelerometers, gyroscopes, and magnetometers to detect events. However, only two out of 27 trialssuccessfully triggered incident detection. The study revealed limitations in the current application of smartwatches for agricultural safety: existing algorithms may not accommodate the complexities of farm-related incidents, limited cellular coverage in rural areas poses a challenge for timely alerts, and differences in sensor performance affect accuracy across devices. To overcome these barriers, the study emphasizes the need for improved fall detection algorithms, enhanced testing methods, and stronger collaboration between researchers and device manufacturers.

LLM-Enhanced VR for Safety Training

Hussain et al. (2024) developed a conversational AI-based virtual reality (VR) system to enhance construction safety training. The system integrates an LLM-powered chatbot with immersive VR to simulate realistic site scenarios and provide interactive, voice-based feedback. Results showed improved engagement and user satisfaction compared to traditional methods, suggesting LLMs can play a key role in creating adaptive, user-centered safety training.

VR Headsets for Safety Education

Schuelke et al. (2025) developed and evaluated an Immersive Virtual Reality (IVR) Ag Safety game, distributed through a mail-out program to secondary school agricultural education programs. Results showed that most of the students found the game engaging and reported increased safety understanding, while faculty supported IVR as an effective teaching tool. Findings suggest IVR is a feasible and scalable method for enhancing agricultural safety education.

Computer Vision in Occupational Safety

Computer vision, a subfield of AI, enables machines to interpret and process visual information from the surrounding environment, such as images and videos. By leveraging advanced algorithms and computational power, computer vision systems mimic human visual perception, allowing them to detect, classify, and monitor objects, events, and patterns (Tian et al., 2020). In the context of agriculture, computer vision holds immense potential to enhance safety by identifying risks, automating hazard detection, and improving decision-making processes. Agricultural environments are inherently dynamic and often hazardous, with workers exposed to risks from heavy machinery, livestock, uneven terrain, and chemical usage. Computer vision systems can process vast amounts of visual data in real time or near real time, enabling the rapid identification of safety concerns (Guo et al., 2021). More broadly, in industrial and occupational settings, computer vision has been applied for fall detection, PPE compliance monitoring, and unsafe behavior recognition (Fang et al., 2020; Zaidi et al., 2024b), offering scalable, non-intrusive solutions that augment traditional safety protocols and reduce workplace injuries.

Application of Computer Vision in Occupational Safety

Using Cameras and Object Detection Models

Computer vision systems have been used in occupational safety to monitor PPE compliance and detect unsafe behaviors on construction sites, enhancing proactive risk management (Zaidi et al., 2024b). In agricultural safety, a key application of computer vision is reducing injuries related to tractors and other machinery through camera-based object detection systems. Modern tractors and autonomous vehicles now use multi-camera setups and algorithms like YOLO and Faster R-CNN to detect humans and automatically stop or reroute when a person is nearby—crucial in low-visibility conditions such as fog or nighttime (Chen and Noguchi, 2023; Redmon et al., 2016; Ren et al., 2016). Detection performance can be further improved by integrating reflective clothing or wearables that augment the camera’s ability to identify workers in non-ideal lighting conditions. These systems also advance autonomous machinery by enabling real-time detection and response (Aby and Issa, 2023). Jung et al. (2020) developed a four-camera YOLO-v3-based recognition system for autonomous tractors, achieving 88.43% precision and 86.19% recall in challenging farm scenarios, effectively preventing accidents.

Drones and Computer Vision

Drones, also known as unmanned aerial vehicles (UAVs), have become an essential tool in modern agriculture, providing real-time aerial surveillance, monitoring, and automated decision-making capabilities. These aerial systems are equipped with high-resolution cameras and advanced sensors, enabling them to capture visual data from farm environments (Akbari et al., 2021). The integration of computer vision allows drones to analyze images and videos, detecting potential hazards and improving overall safety. By leveraging deep learning algorithms, such as convolutional neural networks (CNNs), drones can autonomously identify obstacles, hazardous zones, or workers in the field (Munawar et al., 2022).

Drones have been deployed in occupational safety for tasks like fall detection, hazardous site inspection, and PPE compliance monitoring, offering a safer and more scalable alternative to manual surveillance (Mohsan et al., 2023; Akinsemoyin et al., 2023). By integrating autonomous navigation and advanced computer vision, drones contribute significantly to both operational efficiency and worker protection across high-risk industries. Drones can help prevent accidents by using computer vision-based obstacle detection to navigate around machinery, structures, and workers (Howard et al., 2017). Thermal imaging and object detection capabilities further allow drones to locate individuals in distress on large farms, improving emergency response times (Park and Yeom, 2021). One key aspect of drone-based safety systems isobstacle detection and collision avoidance, which is crucial for ensuring the safe operation of UAVs in complex agricultural environments. Due to the limitations of heavy sensors like radar, lightweight monocular cameras are often employed to detect obstacles and analyze depth by constructing a three-dimensional representation of the environment (Aswini and Uma, 2018). This capability is particularly valuable where monitoring safety risks manually is challenging.

Discussion

In this discussion, the common applications and challenges of emerging AI techniques in the context of occupational safety were examined. It begins by exploring the use of predictive models, highlighting their potential, challenges and common evaluation methods in occupational safety. Similarly, the challenges of applying and evaluating LLMs in occupational safety were discussed. The discussion also considers wearable devices, emphasizing the practical barrier, which is worker acceptance. Lastly, the role of computer vision in safety monitoring was reviewed, along with the foundational models that enable visual detection of hazards in real-time settings.

Predictive Models

Significance of Feature Selection

In ML predictive modeling, feature selection is essential as it directly influences a model’s accuracy and interpretability, allowing it to concentrate on the most relevant factors in occupational incidents, thereby reducing noise and enhancing generalization (Kakhki et al., 2020a). Based on reviewed articles, feature engineering plays a critical role in finding risk factors and transforming raw data into meaningful predictors that capture the core information embedded in the inputs. Common methods for feature selection include correlation analysis, mutual information, recursive feature elimination, and regularization techniques (Fonti and Belitser, 2017). Additionally, model-based feature importance methods, such as those in random forests or gradient boosting, help pinpoint the most influential factors in incident occurrence (Louppe et al., 2013; Chen and Guestrin, 2016). While identifying the primary determinants of an incident offers valuable interpretive insight into safety management, traditional methods fall short when the risk arises from complex nonlinear interactions among contributing factors. Nevertheless, isolating factors that have the strongest influence makes it easier for safety professionals to interpret and act upon an incident (Guyon and Elisseeff, 2003).

Rare Events in Occupational Safety

In occupational safety, incidents are usually rare, making binary classification a key area of focus. ML predictive models face unique challenges for rare event prediction due to class imbalance, where the number of safe (negative) cases extremely outweighs the hazardous (positive) cases (Kakhki et al., 2019a). Addressing these issues involves several strategies which are explained in the next section.

Techniques for Handling Class Imbalance

Evaluation Metrics for Predictive Models

In safety-critical domains like agriculture, evaluating predictive models requires a focus on metrics that reflect the model's effectiveness in detecting incidents accurately. The following metrics are frequently used (Zhou et al., 2020):

Large Language Models

Prompt Engineering

Fine-tuning and prompt engineering are two key strategies for effectively leveraging LLMs in occupational safety applications. Fine-tuning LLMs on domain-specific occupational safety data, such as incident reports, injury logs, and official guidelines, enables the models to internalize sector-specific hazards and protocols, thereby enhancing the accuracy and relevance of their outputs. Prompt engineering, the deliberate design of input queries, further improves performance by steering the LLM to generate safety checklists, translate regulations, or organize injury reports. (Vatsal and Dubey, 2024; Zhang et al., 2024). This is particularly important as the vast repository of safety documents contains valuable safety insights often inaccessible to practitioners. To ensure reliability, LLM outputs should undergo systematic evaluation against real-world scenarios, benchmark safety standards, and expert feedback. Retrieval-augmented generation (RAG) though nascent in agricultural safety applications holds considerable promise for improving output accuracy by reducing hallucinations.

Evaluation

Evaluating these models’ performance depends on the task at hand. For classification or question-answering, accuracy stays a standard metric (Guo et al., 2023), while F1 score provides a balanced measure of precision and recall, particularly in entity recognition and other token-level tasks. For summarization and content extraction, ROUGE (Recall-Oriented Understudy for Gisting Evaluation) assesses overlap with reference summaries via n-gram comparison. However, for generative and open-ended outputs, human evaluation remains essential, judging output quality based on fluency, coherence, relevance, and informativeness (Elangovan et al., 2024). As LLMs are applied in sensitive domains like occupational and agricultural safety, reliable evaluation methods must blend both quantitative benchmarks and expert human judgment to ensure safe and meaningful deployments. Ensuring the accuracy and reliability of LLM-generated outputs is crucial, especially in safety-critical applications.

Wearable Devices

Four types of wearable devices with significant potential to reduce injury risks in occupational safety were discussed: heat stress monitoring devices,exoskeletons, wristbands, and VR. Each of these solutions addresses specific safety challenges and offers innovative ways to enhance the well-being of agricultural workers. Exoskeletons hold promise in reducing the risk of WMSDs for agricultural workers performing repetitive or physically taxing tasks such as lifting, bending, and carrying loads. For instance, passive back-supporting exoskeletons have been tested on farms to evaluate their impact on productivity, health, and safety. Farmers reported benefits such as reduced strain and enhanced task efficiency, highlighting the potential for exoskeletons to enhance both safety and productivity (Thamsuwan et al., 2020). Moreover, wearable sensors enable continuous and objective assessment of ergonomic risks using frameworks like RULA, REBA, and Liberty Mutual indices. This is a capability that was largely impractical with traditional observational approaches. However, the effectiveness of these wearable systems also depends on reliable connectivity, which remains a challenge in isolated rural agricultural environments. Similarly, although AI-augmented VR training can increase user engagement, further research is needed to assess whether this engagement translates to a measurable difference in training effectiveness.

Worker’s Acceptance of Wearable Devices: Safety Context and Challenges

The acceptance of wearable safety devices among occupational workers is influenced by several factors, including perceived safety, comfort, ease of use, and task compatibility. Studies have shown that passive exoskeletons are considered acceptable for certain physical tasks, with potential to reduce fatigue and injury risk while improving productivity (Omoniyi et al., 2020). However, challenges remain regarding their practical deployment across diverse occupational settings. Key concerns include ensuring inclusive design, particularly for women, older workers, and individuals with varying physical capabilities, along with accommodating cultural perceptions, task-specific usability, and long-duration comfort (Upasani et al., 2019). Additionally, social acceptance and workplace culture play crucial roles in influencing adoption. Advancements in AI-driven hybrid and active exoskeleton systems are expected to enhance functionality and personalization, making them more adaptable to the dynamic needs of workers (Jakob et al., 2023).

Computer Vision

Over the past decade, computer vision has undergone transformative changes driven by deep learning innovations. The introduction of convolutional neural networks (CNNs), beginning with AlexNet (Krizhevsky et al., 2012) and followed by deeper architectures such as ResNet (He et al., 2016), has substantially improved machines’ ability to extract hierarchical features from images. These models were pre-trained on large-scale datasets like ImageNet, which contains over a million labeled images (Deng et al., 2009), enabling transfer learning for diverse tasks. Subsequent advances in object detection—such as R-CNN and its faster variants (Girshick et al., 2014; Ren et al., 2016)—and single-shot detectors like YOLO (Redmon et al., 2016) and SSD (Liu et al., 2016) have pushed detection accuracy and inference speed to new heights. On the segmentation front, fully convolutional networks (Long et al., 2015) and Mask R-CNN (He et al., 2017) have enabled pixel-level classification in real time. Together, these breakthroughs have laid a robust foundation for applying computer vision to specialized domains, including occupational safety.

Recent progress in object detection and segmentation models in the computer vision fields has motivated researchers to make use of these advancements to improve safety in occupational risky fields. High-resolution cameras and drones equipped with thermal, RGB, and multispectral imaging capabilities have expanded the scope of data collection in agricultural settings. Meanwhile, breakthroughs in algorithms, such as convolutional neural networks (CNNs), have revolutionized the ability of machines to analyze visual data with remarkable accuracy. In this method, models previously trained on millions of images (e.g., the ImageNet dataset; Deng et al., 2009) are leveraged for downstream tasks. These technological advancements have paved the way for computer vision to become a cornerstone of occupational safety, offering innovative solutions to long-standing challenges in risk management and hazard prevention. As these capabilities are increasingly embedded in agricultural machinery, the nature of human-machine interaction is also evolving in ways that introduce new safety challenges.

Conclusions

Artificial intelligence has demonstrated significant potential in enhancing occupational safety by offering advanced tools for injury prediction and real-time risk assessment. This review examined the application of AI techniques, including predictive modeling, LLMs, computer vision, and wearable technologies, to mitigate risk to workers. The findings suggest that AI-driven predictive models can analyze historical data to identify potential hazards, LLMs improve injury report analysis, data creation, and knowledge extraction, computer vision and remote sensing technologies enhance environmental monitoring, and wearable devices provide real-time health and safety tracking for workers. Despite these advancements, the integration of AI in occupational safety remains in its early stages, with limited research directly focusing on this application. Many of the techniques explored in this review have been adapted from high-risk industries, such as construction and agricultural safety, demonstrating the interdisciplinary nature of AI-driven safety solutions. Future research should focus on improving the accuracy and interpretability of AI models, addressing data limitations through better data collection and standardization, and ensuring the practical deployment of AI-driven safety interventions in real-world settings. Collaboration between researchers, policymakers, and stakeholders will be essential to developing AI solutions that are both effective and accessible to workers. By leveraging the capabilities of AI, the industries can move toward a safer and more resilient future and reducing injury risks.

Acknowledgments

This project was partially supported by the Centers for Disease Control and Prevention–National Institute for Occupational Safety and Health through a grant to the University of Illinois at Chicago (Grant No. T42/OH008672), with additional partial faculty research support provided through the USDA National Institute of Food and Agriculture, Hatch Project 7007373, and the University of Wisconsin–Madison Agricultural Experiment Station.

References

Abdulsadig, R. S., & Rodriguez-Villegas, E. (2024). A comparative study in class imbalance mitigation when working with physiological signals. Front. Digit. Health, 6. https://doi.org/10.3389/fdgth.2024.1377165

Abhishek, K., & Abdelaziz, M. (2023). Machine learning for imbalanced data: Tackle imbalanced datasets using machine learning and deep learning techniques. Packt Publishing.

Aby, G. R., & Issa, S. F. (2023). Safety of automated agricultural machineries: A systematic literature review. Safety, 9(1), 13. https://doi.org/10.3390/safety9010013

Agostinelli, T., Generosi, A., Ceccacci, S., & Mengoni, M. (2024). Validation of computer vision-based ergonomic risk assessment tools for real manufacturing environments. Sci. Rep., 14(1), 27785. https://doi.org/10.1038/s41598-024-79373-4

Ahmadi, E., Muley, S., & Wang, C. (2025). Automatic construction accident report analysis using large language models (LLMs). J. Intell. Constr., 3(1), 1-10. https://doi.org/10.26599/JIC.2024.9180039

Akbari, Y., Almaadeed, N., Al-maadeed, S., & Elharrouss, O. (2021). Applications, databases and open computer vision research from drone videos and images: A survey. Artif. Intell. Rev., 54(5), 3887-3938. https://doi.org/10.1007/s10462-020-09943-1

Akinsemoyin, A., Awolusi, I., Chakraborty, D., Al-Bayati, A. J., & Akanmu, A. (2023). Unmanned Aerial Systems and deep learning for safety and health activity monitoring on construction sites. Sensors, 23(15), 6690. https://doi.org/10.3390/s23156690

Arachchige, S. D., Piyathilaka, L., Sul, J.-H., & Preethichandra, D. M. (2024). A review of potential exoskeletons for the prevention of work-related musculoskeletal disorders in agriculture. Sensors, 24(21), 7026. https://doi.org/10.3390/s24217026

Aswini, N., & Uma, S. V. (2018, September). Obstacle detection in drones using computer vision algorithm. In International Symposium on Signal Processing and Intelligent Recognition Systems (pp. 104-114). Singapore: Springer Singapore.

Badarinath, P. V., Chierichetti, M., & Kakhki, F. D. (2021). A machine learning approach as a surrogate for a finite element analysis: Status of research and application to one dimensional systems. Sensors, 21(5), 1654. https://doi.org/10.3390/s21051654

Borisov, A. V., Borisova, V. L., Konchina, L. V., Kulikova, M. G., & Maslova, K. S. (2021). Application of active exoskeletons with touch sensing, electric drives, energy recuperators and gravity compensators in agricultural technologies. IOP Conf. Ser.: Earth and Environ. Sci., 699(1), 012019. https://doi.org/10.1088/1755-1315/699/1/012019

Breiman, L. (2001). Random forests. Mach. Learn., 45(1), 5-32. https://doi.org/10.1023/A:1010933404324

Bureau of Labor Statistics. (2023). Fatal injuries to agricultural workers in 2021 were the second-lowest in a decade. U.S. Department of Labor. Retrieved from https://www.bls.gov/opub/ted/2023/fatal-injuries-to-agricultural-workers-in-2021-were-the-second-lowest-in-a-decade.htm

Chen, S., & Noguchi, N. (2023). Remote safety system for a robot tractor using a monocular camera and a YOLO-based method. Comput. Electron. Agric., 215, 108409. https://doi.org/10.1016/j.compag.2023.108409

Chen, T., & Guestrin, C. (2016). XGBoost: A scalable tree boosting system. Proc. 22nd ACM SIGKDD Int. Conf. on Knowledge Discovery and Data Mining (pp. 785–794). Association for Computing Machinery. https://doi.org/10.1145/2939672.2939785

Choi, B., Park, J., & Kim, J. H. (2024). Assessment of an arm-support exoskeleton on physical demands, task performance, and usability during simulated agricultural tasks. Int. J. Ind. Ergon., 101, 103569. https://doi.org/10.1016/j.ergon.2024.103569

Choi, J., Gu, B., Chin, S., & Lee, J.-S. (2020). Machine learning predictive model based on national data for fatal accidents of construction workers. Autom. Constr., 110, 102974. https://doi.org/10.1016/j.autcon.2019.102974

Dablain, D., Krawczyk, B., & Chawla, N. V. (2022). DeepSMOTE: Fusing deep learning and SMOTE for imbalanced data. IEEE Trans. Neural Netw. Learn. Syst., 34(9), 6390-6404. https://doi.org/10.1109/TNNLS.2021.3136503

Deng, J., Dong, W., Socher, R., Li, L. J., Li, K., & Li, F.-F. (2009). ImageNet: A large-scale hierarchical image database. Proc. 2009 IEEE Conf. on Computer Vision and Pattern Recognition, (pp. 248-255). https://doi.org/10.1109/CVPR.2009.5206848

Elangovan, A., Liu, L., Xu, L., Bodapati, S. B., & Roth, D. (2024). ConSiDERS-the-human evaluation framework: Rethinking human evaluation for generative large language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (pp. 1137-1160). https://doi.org/ 10.18653/v1/2024.acl-long.63

Etienne, A. J., Field, W. E., Ehlers, S. G., Tormoehlen, R., & Haslett, N. J. (2024). Testing the feasibility of selected, commercially available wearable devices in detecting agricultural-related incidents. J. Agric. Saf. Health, 30(4), 181-204. https://doi.org/10.13031/jash.15985

Fang, W., Ding, L., Love, P. E., Luo, H., Li, H., Peña-Mora, F., ... & Zhou, C. (2020). Computer vision applications in construction safety assurance. Autom. Constr., 110, 103013. https://doi.org/10.1016/j.autcon.2019.103013

Firth, D. (1993). Bias reduction of maximum likelihood estimates. Biometrika, 80(1), 27-38. https://doi.org/10.1093/biomet/80.1.27

Fonti, V., & Belitser, E. (2017). Feature selection using lasso. VU Amsterdam research paper in business analytics, 30, 1-25

Gibson, M. (2012). Food security — A commentary: What is it and why is it so complicated? Foods, 1(1), 18-27. https://doi.org/10.3390/foods1010018

Girshick, R., Donahue, J., Darrell, T., & Malik, J. (2014). Rich feature hierarchies for accurate object detection and semantic segmentation. Proc. 2014 IEEE Conf. on Computer Vision and Pattern Recognition, (pp. 580-587). https://doi.org/10.1109/CVPR.2014.81

Goel, H., Narasimhan, S. S., Akcin, O., & Chinchali, S. (2024). SynDiff-AD: Improving semantic segmentation and end-to-end autonomous driving with synthetic data from latent diffusion models. arXiv preprint arXiv:2411.16776. https://doi.org/10.48550/arXiv.2411.16776

Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S.,... Bengio, Y. (2020). Generative adversarial networks. Commun. ACM, 63(11), 139–144. https://doi.org/10.1145/3422622

Guo, B. H., Zou, Y., Fang, Y., Goh, Y. M., & Zou, P. X. (2021). Computer vision technologies for safety science and management in construction: A critical review and future research directions. Saf. Sci., 135, 105130. https://doi.org/10.1016/j.ssci.2020.105130

Guo, Z., Jin, R., Liu, C., Huang, Y., Shi, D., Yu, L., ... & Xiong, D. (2023). Evaluating large language models: A comprehensive survey. arXiv preprint arXiv:2310.19736.

Guyon, I., & Elisseeff, A. (2003). An introduction to variable and feature selection. J. Mach. Learn. Res., 3(2003), 1157-1182.

He, K., Gkioxari, G., Dollár, P., & Girshick, R. (2017). Mask R-CNN. Prcoc. 2017 IEEE Int. Conf. on Computer Vision (ICCV), (pp. 2980-2988). https://doi.org/10.1109/ICCV.2017.322

He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. Proc. 2016 IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), (pp. 770-778). https://doi.org/10.1109/CVPR.2016.90

Howard, J., Murashov, V., & Branche, C. (2017, October 23). Can drones make construction safer? NIOSH Science Blogs. Centers for Disease Control and Prevention. https://www.cdc.gov/niosh/blogs/2017/drones.html

Hussain, R., Sabir, A., Lee, D.-Y., Zaidi, S. F., Pedro, A., Abbas, M. S., & Park, C. (2024). Conversational AI-based VR system to improve construction safety training of migrant workers. Autom. Constr., 160, 105315. https://doi.org/10.1016/j.autcon.2024.105315

Issa, S. F., Cheng, Y. H., & Field, W. (2016). Summary of agricultural confined-space related cases: 1964-2013. Journal of agricultural safety and health, 22(1), 33-45.

Issa, S. F., Patrick, K., Thomson, S., & Rein, B. (2019). Estimating the number of agricultural fatal injuries prevented by agricultural engineering developments in the United States. Safety, 5(4), 63. https://doi.org/10.3390/safety5040063

Jain, H., Dhupper, R., Shrivastava, A., Kumar, D., & Kumari, M. (2023). Leveraging machine learning algorithms for improved disaster preparedness and response through accurate weather pattern and natural disaster prediction. Front. Environ. Sci., 11, 1194918. https://doi.org/10.3389/fenvs.2023.1194918

Jakob, M., Balaguier, R., Park, H., & Trask, C. (2023). Addressing exoskeleton implementation challenges: Case studies of non-acceptance in agriculture. J. Agromed., 28(4), 784-796. https://doi.org/10.1080/1059924X.2023.2236605

Jawahar, G., Sagot, B., & Seddah, D. (2019). What does BERT learn about the structure of language? Proc. 57th Annual Meeting of the Association for Computational Linguistics. https://doi.org/10.18653/v1/P19-1356

Johnson, E., & Wilson, N. (2024). Enhancing Agricultural Machinery Management through Advanced LLM Integration. arXiv preprint arXiv:2407.20588.

Jung, T.-H., Cates, B., Choi, I.-K., Lee, S.-H., & Choi, J.-M. (2020). Multi-camera-based person recognition system for autonomous tractors. Designs, 4(4), 54. https://doi.org/10.3390/designs4040054

Kakhki, F.D., Freeman, S. A., & Mosher, G. A. (2020a). Applied machine learning in agro-manufacturing occupational incidents. Procedia Manuf., 48, 24–30. https://doi.org/10.1016/j.promfg.2020.05.016

Kakhki, F.D., Freeman, S. A., & Mosher, G. A. (2019a). Use of neural networks to identify safety prevention priorities in agro-manufacturing operations within commercial grain elevators. Appl. Sci., 9(21), 4690. https://doi.org/10.3390/app9214690

Kakhki, F. D., Freeman, S. A., & Mosher, G. A. (2019b). Evaluating machine learning performance in predicting injury severity in agribusiness industries. Saf. Sci., 117, 257-262. https://doi.org/10.1016/j.ssci.2019.04.026

Kakhki, F. D., Freeman, S. A., & Mosher, G. A. (2020b). Utilization of machine learning in analyzing post-incident state of occupational injuries in agro-manufacturing industries. In P. M. Arezes, & R. L. Boring (Ed.), Advances in Safety Management and Human Performance - Proc. AHFE 2020 Virtual Conf. on Safety Management and Human Factors, and Human Error, Reliability, Resilience, and Performance (pp. 3-9). Springer. https://doi.org/10.1007/978-3-030-50946-0_1

Koc, K., Ekmekcioglu, Ö., & Gurgun, A. P. (2023). Developing a national data-driven construction safety management framework with interpretable fatal accident prediction. J. Constr. Eng. Manag., 149(4), 04023010. https://doi.org/10.1061/JCEMD4.COENG-12848

Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). ImageNet classification with deep convolutional neural networks. In F. Pereira, C. J. Burges, L. Bottou, & K. Q. Weinberger (Ed.), Proc. Advances in Neural Information Processing Systems 25 (NIPS 2012), (pp. 1097-1105).

Lehtola, M. M., Rautiainen, R. H., Day, L. M., Schonstein, E., Suutarinen, J., Salminen, S., & Verbeek, J. H. (2008). Effectiveness of interventions in preventing injuries in agriculture — a systematic review and meta-analysis. Scand. J. Work Environ. Health, 34(5), 327-336. https://doi.org/10.5271/sjweh.1279

Li, C. (2022). Ecological Agriculture Safety Supervision System Based on Random Forest Algorithm. In Proc. 2022 Second Int. Conf. on Advanced Technologies in Intelligent Control, Environment, Computing & Communication Engineering (ICATIECE) (pp. 1-5). IEEE. https://doi.org/ 10.1109/ICATIECE56365.2022.10047679

Li, S., Raza, M. M., & Issa, S. (2024). Agricultural injury surveillance in the United States and Canada: A systematic literature review. J. Agromed., 29(2), 122-135. https://doi.org/10.1080/1059924X.2024.2304699

Lincoln, J., Gorucu, S., Khorsandi, F., Aby, G. R., Elliott, K. C., Shutske, J., & Issa, S. F. (2025). Occupational safety research needs in the field of robotics and autonomous machines in agriculture. J. Agric. Saf. Health, 31(3), 217–230. https://doi.org/10.13031/jash.16216

Liu, J., Wong, Z. S., So, H. Y., & Tsui, K. L. (2021). Evaluating resampling methods and structured features to improve fall incident report identification by the severity level. J. Am. Med. Inform. Assoc., 28(8), 1756-1764. https://doi.org/10.1093/jamia/ocab048

Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu, C.-Y., & Berg, A. C. (2016). SSD: Single shot multibox detector. In B. Leibe, J. Matas, N. Sebe, & M. Welling (Ed.), Proc. Computer Vision – ECCV 2016, 14th European Conference (pp. 21-37). Springer. https://doi.org/10.1007/978-3-319-46448-0_2

Long, J., Shelhamer, E., & Darrell, T. (2015). Fully convolutional networks for semantic segmentation. Proc. 2015 IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), (pp. 3431-3440). https://doi.org/10.1109/CVPR.2015.7298965

Louppe, G., Wehenkel, L., Sutera, A., & Geurts, P. (2013). Understanding variable importances in forests of randomized trees. Advances in Neural Information Processing Systems, 26.

Lu, L., Zhang, J., Xie, Y., Gao, F., Xu, S., Wu, X., & Ye, Z. (2020). Wearable health devices in health care: Narrative systematic review. JMIR Mhealth Uhealth, 8(11), e18907. https://doi.org/10.2196/18907

Luque, A., Carrasco, A., Martín, A., & de las Heras, A. (2019). The impact of class imbalance in classification performance metrics based on the binary confusion matrix. Pattern Recognit., 91, 216-231. https://doi.org/10.1016/j.patcog.2019.02.023

Massardi, S., Pinto-Fernandez, D., Babic, J., Dežman, M., Trošt, A., Grosu, V.,... Torricelli, D. (2023). Relevance of hazards in exoskeleton applications: A survey-based enquiry. J. NeuroEng. Rehabil., 20(1), 68. https://doi.org/10.1186/s12984-023-01191-y

McCurdy, S. A., & Carroll, D. J. (2000). Agricultural injury. Am. J. Ind. Med., 38(4), 463-480. https://doi.org/10.1002/1097-0274(200010)38:4<463::AID-AJIM13>3.0.CO;2-N

Min, B., Ross, H., Sulem, E., Veyseh, A. P., Nguyen, T. H., Sainz, O.,... Roth, D. (2023). Recent advances in natural language processing via large pre-trained language models: A survey. ACM Comput. Surv., 56(2), Article 30. https://doi.org/10.1145/3605943

Mohsan, S. A. H., Othman, N. Q. H., Li, Y., Alsharif, M. H., & Khan, M. A. (2023). Unmanned aerial vehicles (UAVs): Practical aspects, applications, open challenges, security issues, and future trends. Intell. Serv. Robotics, 16(1), 109-137. https://doi.org/10.1007/s11370-022-00452-4

Muller, J. M., Petti, D. J., Li, C., Gorucu, S., Pilz, M., & Weichelt, B. (2024). Investigating the use of large language models in agricultural injury surveillance. Paper No. 2400572. Proc. 2024 ASABE Annual Int. Meeting. St. Joseph, MI: ASABE. https://doi.org/10.13031/aim.202400572

Munawar, H. S., Ullah, F., Heravi, A., Thaheem, M. J., & Maqsoom, A. (2022). Inspecting buildings using drones and computer vision: A machine learning approach to detect cracks and damages. Drones, 6(1), 5. https://doi.org/10.3390/drones6010005

Nadas, M., Diosan, L., & Tomescu, A. (2025). Synthetic data generation using large language models: Advances in text and code. IEEE Access, 13, 134615-134633. https://doi.org/10.1109/ACCESS.2025.3589503

Nguyen, T.-H.-Y., Bertin, M., Bodin, J., Fouquet, N., Bonvallot, N., & Roquelaure, Y. (2018). Multiple exposures and coexposures to occupational hazards among agricultural workers: A systematic review of observational studies. Saf. Health Work, 9(3), 239-248. https://doi.org/10.1016/j.shaw.2018.04.002

Northcutt, C. G., Athalye, A., & Mueller, J. (2021). Pervasive label errors in test sets destabilize machine learning benchmarks. arXiv preprint arXiv:2103.14749. https://doi.org/10.48550/arXiv.2103.14749

Obuchowski, N. A. (1997). Nonparametric analysis of clustered ROC curve data. Biometrics, 53(2), 567-578. https://doi.org/10.2307/2533958

Occupational Safety and Health Administration (OSHA). (2025). Agricultural operations. United States Department of Labor. Retrieved from https://www.osha.gov/agriculture

Ogundimu, E. O. (2019). Prediction of default probability by using statistical models for rare events. J. R. Stat. Soc. Ser. A Stat. Soc., 182(4), 1143-1162. https://doi.org/10.1111/rssa.12467

Ojha, A., Gautam, Y., Jebelli, H., & Akanmu, A. (2024). Physiological impact of powered back-support exoskeletons in construction: Analyzing muscle fatigue, metabolic cost, ergonomic risks, and stability. Autom. Constr., 168, 105742. https://doi.org/10.1016/j.autcon.2024.105742

Omoniyi, A., Trask, C., Milosavljevic, S., & Thamsuwan, O. (2020). Farmers’ perceptions of exoskeleton use on farms: Finding the right tool for the work(er). Int. J. Ind. Ergon., 80, 103036. https://doi.org/10.1016/j.ergon.2020.103036

Pallathadka, H., Mustafa, M., Sanchez, D. T., Sekhar Sajja, G., Gour, S., & Naved, M. (2023). Impact of machine learning on management, healthcare and agriculture. Mater. Today: Proc., 80(Part 3), 2803-2806. https://doi.org/10.1016/j.matpr.2021.07.042

Park, D.-M., & Yeom, S. (2021). Human detection and posture estimation with IR thermal images by a drone. Proc. 2021 4th Artificial Intelligence and Cloud Computing Conference (pp. 115–121). Association for Computing Machinery. https://doi.org/10.1145/3508259.3508276

Patel, V., Chesmore, A., Legner, C. M., & Pandey, S. (2022). Trends in workplace wearable technologies and connected-worker solutions for next-generation occupational safety, health, and productivity. arXiv preprint arXiv:2205.11740, 4(1). https://doi.org/10.48550/arXiv.2205.11740

Pishgar, M., Issa, S. F., Sietsema, M., Pratap, P., & Darabi, H. (2021). REDECA: A novel framework to review artificial intelligence and its applications in occupational safety and health. Int. J. Environ. Res. Public. Health, 18(13), 6705. https://doi.org/10.3390/ijerph18136705

Probst, T. M., Bettac, E. L., & Austin, C. T. (2019). Accident under-reporting in the workplace (pp. 30–47). In R. J. Burke & A. M. Richardsen (Eds.), Increasing occupational health and safety in workplaces (Business 2019). Edward Elgar Publishing. https://doi.org/10.4337/9781788118095.00009

Rathod, S. B., Mahajan, R. A., Khadkikar, P. A., Vyawahare, H. R., & Patil, P. R. (2024). Improving workplace safety with AI-powered predictive analytics: Enhancing workplace security. In S. Ponnusamy, V. Bora, P. M. Daigavane, & S. S. Wazalwar (Eds.), AI tools and applications for women’s safety (pp. 232-249). Hershey, PA: IGI Global Scientific Publishing. https://doi.org/10.4018/979-8-3693-1435-7.ch014

Raza, M. M., Li, S., & Issa, S. F. (2024). Global patterns of agricultural machine and equipment injuries - A systematic literature review. J. Agromed., 29(2), 214-234. https://doi.org/10.1080/1059924X.2024.2304704

Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2016). You Only Look Once: Unified, real-time object detection. Proc. 2016 IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), (pp. 779-788). https://doi.org/10.1109/CVPR.2016.91

Reeder, B., & David, A. (2016). Health at hand: A systematic review of smart watch uses for health and wellness. J. Biomed. Inform., 63, 269-276. https://doi.org/10.1016/j.jbi.2016.09.001

Ren, S., He, K., Girshick, R., & Sun, J. (2016). Faster R-CNN: Towards real-time object detection with region proposal networks. IEEE Trans. Pattern Anal. Mach. Intell., 39(6), 1137-1149. https://doi.org/10.1109/TPAMI.2016.2577031

Rodríguez, H., Ekström, M., Ramirez, B., & Marback, P. (2024). Novel semi-active upper-body exoskeleton for agriculture applications. Proc. 2024 9th Int. Engineering, Sciences and Technology Conference (IESTEC), (pp. 630-635). https://doi.org/10.1109/IESTEC62784.2024.10820212

Roumeliotis, K. I., & Tselikas, N. D. (2023). ChatGPT and Open-AI models: A preliminary review. Future Internet, 15(6), 192. https://doi.org/10.3390/fi15060192

Schuelke, S., Yoder, A., Kreifels, M., & Kupzyk, K. (2025). Implementation of a statewide youth ag safety immersive virtual reality program. J. Agromed., 30(3), 496-505. https://doi.org/10.1080/1059924X.2025.2473456

Sengül, G., Karakaya, M., Misra, S., Abayomi-Alli, O. O., & Damaševicius, R. (2022). Deep learning based fall detection using smartwatches for healthcare applications. Biomed. Signal Process. Control, 71(Part B), 103242. https://doi.org/10.1016/j.bspc.2021.103242

Shakerian, S., Habibnezhad, M., Ojha, A., Lee, G., Liu, Y., Jebelli, H., & Lee, S. (2021). Assessing occupational risk of heat stress at construction: A worker-centric wearable sensor-based approach. Saf. Sci., 142, 105395. https://doi.org/10.1016/j.ssci.2021.105395

Shutske, J. M. (2023). Editorial: Harnessing the power of large language models in agricultural safety & health. J. Agric. Saf. Health, 29(4), 205-224. https://doi.org/10.13031/jash.15841

Shutske, J. M., Issa, S. F., Johnson, T., Khorsandi, F., Pate, M. L., Gorucu, S., … Versweyveld, J. (2025a). SAFER AG – Risk assessment, data, design standards, and regulation: Needs and recommendations. J. Agric. Saf. Health, 31(1), 1–13. https://doi.org/10.13031/jash.15855

Shutske, J. M., Issa, S. F., Khorsandi, F., Pate, M. L., Gorucu, S., Walsh, J., … Li, S. (2025b). SaferAg – Engineering safety in emerging agricultural machinery: Risk assessment, data needs, and standards. J. Agric. Saf. Health, 31(4), 311–324. https://doi.org/10.13031/jash.16516

Snorkel AI. (2024). Data Labeling: A Complete Guide. Retrieved from https://snorkel.ai/data-labeling/

Thamsuwan, O., Milosavljevic, S., Srinivasan, D., & Trask, C. (2020). Potential exoskeleton uses for reducing low back muscular activity during farm tasks. Am. J. Ind. Med., 63(11), 1017-1028. https://doi.org/10.1002/ajim.23180

Tian, H., Wang, T., Liu, Y., Qiao, X., & Li, Y. (2020). Computer vision technology in agricultural automation — A review. Inf. Process. Agric., 7(1), 1-19. https://doi.org/10.1016/j.inpa.2019.09.006

Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M. A., Lacroix, T.,... Lample, G. (2023). Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971. https://doi.org/10.48550/arXiv.2302.13971

Upasani, S., Franco, R., Niewolny, K., & Srinivasan, D. (2019). The potential for exoskeletons to improve health and safety in agriculture — Perspectives from service providers. IISE Trans. Occup. Ergon. Hum. Factors, 7(3-4), 222-229. https://doi.org/10.1080/24725838.2019.1575930

Vatsal, S., & Dubey, H. (2024). A survey of prompt engineering methods in large language models for different NLP tasks. arXiv preprint arXiv:2407.12994. https://doi.org/10.48550/arXiv.2407.12994

Wagoner, R. S., López-Gálvez, N. I., de Zapien, J. G., Griffin, S. C., Canales, R. A., & Beamer, P. I. (2020). An occupational heat stress and hydration assessment of agricultural workers in North Mexico. Int. J. Environ. Res. Public. Health, 17(6), 2102. https://doi.org/10.3390/ijerph17062102

Weichelt, B. P., Pilz, M., Burke, R., Puthoff, D., & Namkoong, K. (2024). The potential of AI and ChatGPT in improving agricultural injury and illness surveillance programming and dissemination. J. Agromed., 29(2), 150-154. https://doi.org/10.1080/1059924X.2023.2284959

Zaidi, S. F. A., Abbas, M. S., Hussain, R., Sabir, A., Khan, N., Yang, J., & Park, C. (2024a). iSafe chatbot: natural language processing and large language model driven construction safety learning through OSHA rules and video content delivery. In Proc. Int. Conf. on Construction Engineering and Project Management (pp. 1238-1245). Korea Institute of Construction Engineering and Management.

Zaidi, S. F. A., Yang, J., Abbas, M. S., Hussain, R., Lee, D., & Park, C. (2024b). Vision-based construction safety monitoring utilizing temporal analysis to reduce false alarms. Buildings, 14(6), 1878. https://doi.org/10.3390/buildings14061878

Zhang, J., Fiers, P., Witte, K. A., Jackson, R. W., Poggensee, K. L., Atkeson, C. G., & Collins, S. H. (2017). Human-in-the-loop optimization of exoskeleton assistance during walking. Science, 356(6344), 1280-1284. https://doi.org/10.1126/science.aal5054

Zhang, X., Talukdar, N., Vemulapalli, S., Ahn, S., Wang, J., Meng, H.,... Chen, B. (2024). Comparison of prompt engineering and fine-tuning strategies in large language models in the classification of clinical notes. Proc. AMIA Jt. Summits Transl. Sci. Proc., 478–487. https://doi.org/10.1101/2024.02.07.24302444

Zhou, X., Lu, P., Zheng, Z., Tolliver, D., & Keramati, A. (2020). Accident prediction accuracy assessment for highway-rail grade crossings using random forest algorithm compared with decision tree. Reliab. Eng. Syst. Saf., 200, 106931. https://doi.org/10.1016/j.ress.2020.106931

Zhu, J., Shi, Q., Li, Q., Shou, W., Li, H., & Wu, P. (2023). Developing predictive models of construction fatality characteristics using machine learning. Saf. Sci., 164, 106149. https://doi.org/10.1016/j.ssci.2023.106149

Table 1. Summary of reviewed articles applying predictive models in health, safety, agriculture, and agricultural safety. Studies are grouped into two categories: injury prediction and risk factors identification.^[a],[b]
	Article	Focus Area	Feature Selection	Predictive Models	Evaluation Metrics^[a]	Methods to Deal with Imbalance Dataset^[b],[c]
Injury Prediction	(Choi et al., 2020)	Prediction of injuries	Yes, DT	DT,NN, RF, SGBM, SVMLK	Acc, Kappa	NA
	(Kakhki et al., 2020a)	Classify incidents by severity level.	Yes, Chi-square and Bootstrap	NB, RF	Acc, Err, F1-score, R, P, Specificity	NA
	(Kakhki et al., 2019a)	Use of Neural Networks to Identify Safety Prevention Priorities in Agro-Manufacturing Operations within Commercial Grain Elevators	NA	MLP, RBF	AUC	NA
	(Liu et al., 2021)	Improve Fall Incident Severity Reporting	Yes, Permutation importance	SVM, NB, RF	F1- score, R, P	RUS, ROS, Tomek Link, bSMOTE-1, SMOTE, bSMOTE-2, ProWSyn, polynorm-fit
	(Zhu et al., 2023)	Predicting construction fatality characteristics	Yes, DT	DT,NN, RF, SGBM, SVMLK	Acc, Kappa	NA
Risk Factors Identification	(Kakhki et al., 2020b)	Post-incident injury prediction	NA	CHAID, CART	Acc	NA
	(Kakhki et al., 2019b)	Predicting injury	Yes, Chi-square	SVM (linear), SVM (quadratic), SVM (RFB), BT,NB	R, P, F-score, Acc, Err, Specificity	NA
	(Koc et al., 2023)	ML for construction accident fatal prediction	Yes, SHAP	RF, PSO	Acc, F1-score, AUC	NA
	(Ogundimu, 2019)	credit risk prediction for credit card defaults, focusing on rare event classification	NA	LR, Firth’s (1993) method, Firth’s logistic regression with added covariate, log-F prior penalty, Ridge regression, Linear discriminant analysis, Generalized extreme value regression model	Brier score, Calibration slope, AUC, AUPRC	SMOTE and ROSE
^[a] Evaluation metrics: P = Precision. R = Recall. Err = Error. Acc = Accuracy. RF = Random Forest. NB = Naïve Bayes. MLP = Multilayer perceptron. RBF = Radial basis function. ^[b] Machine Learning Methods: polynom-fit = Fit a polynomial function to all minority samples; ProWSyn = Generate synthetic minority data with weighted proximity; B-2 = Identify borderline minority samples and create synthetic points; bSMOTE-1 = Identify borderline samples with more majority neighbors and generate synthetic data; SMOTE = Generate synthetic minority points between a sample and K-nearest neighbors; ROS = Replicate randomly selected minority examples; Tomek Link = Remove majority examples linked to minority; RUS = Randomly remove majority examples; SHAP = SHapley Additive exPlanations.

Table 2. Summary of reviewed articles on application of large language models in different areas of occupational safety and health.
Article	Focus Area	LLM	Evaluation Metrics
(Ahmadi et al., 2025)	Analyzing accident report with LLM	GPT-3.5, GPT-4.0, Gemini Pro, and LLaMA 3.1	Accuracy
(Muller et al., 2024)	LLM in agricultural injury surveillance	ChatGPT 3.5 and 4, Llama 2	Average accuracy
(Weichelt et al., 2024)	LLM in agricultural injury surveillance	ChatGPT	Not specified
(Nada? et al., 2025)	Generating text and code data	GPT-3, 3.5, and 4. LLaMA 3, Claude 3.7, DeepSeek’s R1	Not specified
(Shutske, 2023)	LLM in agricultural safety education	ChatGPT 3.5 and 4	Human evaluation
(Zaidi et al., 2024a)	Developing chatbot for safety related questions	ChatGPT 3.5	Human evaluation

Table 3. Advantages and disadvantages for each exoskeleton type for farmers.
Types	Pros	Cons
Active	Dynamic adaptability, AI-driven controls for natural movement, predictive safety monitoring.	High cost, limited battery life, dependency risks, risk of overconfidence during use.
Passive	Simple maintenance, no reliance on power, unlimited operational time.	Limited adaptability, no real-time adjustments, less effective for variable tasks.
Hybrid	Balanced adaptability and simplicity, reduced cost compared to active.	Moderate power reliance, not as robust as purely active systems.

Table 4. Confusion matrix used to evaluate classification performance by comparing actual and predicted outcomes.
	Predicted Positive	Predicted Negative
Actual Positive	True Positive (TP) Incident Incident	False Negative (FN) Incident No Incident
Actual Negative	False Positive (FP) No Incident Incident	True Negative (TN) No Incident No Incident

Enhancing Occupational Safety Through AI: A Review of Key AI Technologies

Published in Journal of the ASABE 69(1): 165-179 (doi: 10.13031/ja.16502). Copyright 2026 American Society of Agricultural and Biological Engineers.

Materials and Methods

Results

Predictive Models in Occupational Safety

Application of Predictive Models in Occupational Safety

Predicting Injuries and Incidents

Identify Injury Risk Factors

Large Language Models in Occupational Safety

Application of LLMs in Occupational Safety

Incident Reports Analysis

Safety Education

Developing Chatbot

Data Generation and Annotation

Wearable Devices for Human Safety in Occupational Settings

Application of Wearable Devices in Occupational Safety

Heat Stress Monitor

Enhancing Safety and Reducing Work-Related Risks with Exoskeletons

Types of Exoskeletons and the Role of AI

Wristbands for Tracking Health Metrics

LLM-Enhanced VR for Safety Training

VR Headsets for Safety Education

Computer Vision in Occupational Safety

Application of Computer Vision in Occupational Safety

Using Cameras and Object Detection Models

Drones and Computer Vision

Discussion

Predictive Models

Significance of Feature Selection

Rare Events in Occupational Safety

Techniques for Handling Class Imbalance

Evaluation Metrics for Predictive Models

Large Language Models

Prompt Engineering

Evaluation

Wearable Devices

Worker’s Acceptance of Wearable Devices: Safety Context and Challenges

Computer Vision

Conclusions

Acknowledgments

References