Immersive technologies for evaluating industrial safety training in high-risk environments: a review on opportunities and challenges

Evaluating industrial safety training in high-risk environments remains a methodological challenge, especially in sectors such as construction and mining, where the reliable measurement of knowledge transfer and behavioral change is limited. This study aims to identify the key attributes required for a model that leverages immersive technologies to evaluate safety training in high-risk scenarios, thereby advancing occupational safety research and informing the design of industrial training programs. A PRISMA-guided systematic review of 37 peer-reviewed studies (2021–2025) was conducted across Scopus, ScienceDirect and Web of Science, ensuring transparency and reproducibility. The analysis identified methodological patterns, technological features and research gaps. Most studies addressed immediate outcomes (reaction and learning, corresponding to Levels 1 and 2 of Kirkpatrick’s model), while evidence on behavioral change (Level 3) and organizational impact (Level 4) is scarce. Evaluation strategies are also fragmented, with a predominance of self-report questionnaires and limited use of biometric or performance-based metrics. Emerging approaches combine multimodal Human Action Recognition (HAR), biometric sensing (eye-tracking, EEG, heart rate), and behavioral analytics to enable real-time, performance-based assessment. Adaptive, AI-driven and gamified environments are also gaining relevance, by combining biometric feedback with behavioral data to detect and interpret user actions in real time. By consolidating these attributes, this review delineates the essential components of an immersive evaluation framework that advances methodological rigor and supports safer, human-centered industrial training aligned with Industry 4.0 and Society 5.0. Systematic Review Registration https://doi.org/10.17605/OSF.IO/EKWZ9 .