Falling risk analysis at workplaces through an accident data-driven approach based upon hybrid artificial intelligence (AI) techniques

Abstract

This study proposed an accident data-driven approach using hybrid AI techniques for the quantification of falling risks at workplaces. Six machine learning models and one ensemble learning model were deployed for automatic extraction of causal factors. These causal factors were taken as main nodes in the falling risk Bayesian network (FRBN). Data-driven and knowledge-driven methods were combined for structure learning of the FRBN, based upon algorithms of hill climbing and tree augmented naive Bayes firstly and modification of FRBN through incorporation of knowledge. Sensitive causal factors were determined using parameter-based and evidence-based sensitivity analysis approaches. The FRBN was further adopted for forward and backward causal inferences. The accident data-driven approach through hybrid AI techniques contributes to substantial learning from fall-related accidents. Measures would be tailored according to causal inferences within the FRBN, so that the probability of falling risk will be reduced and negative impacts of fall-from-height (FFH) accidents will be decreased.