Machine learning-based Bayesian framework for interval estimate of unsafe-event prediction in construction

Construction safety is a critical concern for industry and academia, and numerous models and algorithms have been developed to predict incidents or accidents to facilitate proactive decision-making. However, previous studies have been limited due to the inability to account for uncertainties because predictions are given as a single value (i.e., Yes or No) and the failure to integrate subjective judgment. To address these limitations, this research proposes a machine learning-based Bayesian framework for predicting construction incidents using interval estimates. This framework combines a state-of-the-art machine-learning algorithm with a binary Bayesian inference model to develop an incident predictor that considers a range of project characteristics and conditions. Notably, this framework also is capable of incorporating historical or subjective judgment through prior selection and outputs the unsafe event prediction as an interval of possibilities, thus accounting for various uncertainties. The efficacy of our framework was demonstrated in a real-life case study, showcasing its practical implications for proactive decision-making and risk management in the construction industry and representing a valuable contribution to the field of construction safety.