Predicting Firefighter Injury and Entrapment in Urban Firefighting Operations: An Investigation Into the Effectiveness of Modified Fire Time Stages and Machine Learning.

Background and Aims: Urban fires pose a significant threat in terms of property damage and potential loss of life. Firefighters play an important role in managing these incidents, so their safety is a top priority for fire departments and emergency responders. The purpose of this study is to predict the probability of firefighter injury and entrapment before dispatching the operational team or the initial stages of the urban firefighting operations. Methods: In this study, we compare the performance of eight machine learning algorithms in predicting the occurrence of firefighter injuries and entrapment during urban fire incidents. We use data from the Fire Department Operations and Management System (FOMS) of Mashhad's city fire and safety services organization. Specifically, we assess the effectiveness of the generated models through five stages, two of which use additional features calculated using the built-in FOMS functions. Results: We found that the Multi-Layer Perceptron and Kernel Naive Bayes algorithms achieved the highest predictive accuracy for the occurrence of firefighter injuries and entrapments, 96.7% and 96.0%, respectively. Conclusions: The generated models can help to reduce the incidence of injuries and entrapments among firefighters by providing the decision-makers with early prediction of the risk factors associated with an ongoing urban firefighting operation.