Artificial Cognition to Predict and Explain the Potential Unsafe Behaviors of Construction Workers

Abstract

Unsafe behavior is considered the primary cause of construction safety accidents. However, the main measures for unsafe behavior management are real-time monitoring and postevent correction, which cannot prevent unsafe behavior. Therefore, this study attempted to construct an artificial cognition approach to predict the potential unsafe behavior of workers and explain why workers engage in unsafe behaviors. First, based on the cognitive model of unsafe behavior, data on workers were collected with a questionnaire, and the cognitive model was validated. Second, the cognitive process of unsafe behaviors was analyzed using latent class analysis, and the cognitive characteristics of four types of unsafe behaviors were obtained. Subsequently, with the cognitive model of unsafe behavior as the input attribute, seven types of algorithms (gradient Boosting, random forest, naïve bayes, back propagation, K-nearest neighbor, logistic regression, and support vector machine) were used to construct artificial cognition to predict the potential unsafe behaviors of workers. The results showed that all seven algorithms performed well for prediction. Thus, artificial cognition that simulates the cognitive process of unsafe behavior is not limited to particular algorithms. Finally, artificial cognition was empirically validated in a construction project. The findings demonstrated that artificial cognition could effectively predict the potential unsafe behavior of workers and provide an explanation for why workers engage in unsafe behaviors.