Development of data-driven influence model to relate the workplace environment to human error

Because human error plays a direct role in accidents, studying the causal relationship between the environment and human error is essential to prevent mishaps. However, these relationships have been explored solely using bivariate statistical analysis and thus require more intermediate factors to emphasize the need for monitoring and controlling human error by improving the workplace environment. Moreover, prevalent studies rely heavily on expert experience, which is subjective and creates potential estimation noise. In this study, the mechanism whereby environmental factors influence behavior and its associate factors is learned with an algorithm using a Bayesian network structure. Rather than being simply data-driven, the algorithm initiates learning from prior knowledge, the theoretical causal chain in the cognitive reliability and error analysis method (CREAM), and revises the learning approach against safety inspection data if necessary. The learned Bayesian network shows that human error and incorrect sequencing result from a combination of limited cognitive functions and improper spatial/workmanship arrangements caused by equipment defects, improper design, and management problems. Bridging the gaps in previous studies, the action interface revealed by this study is useful for on-site quality control.