A new procedure for assessing and predicting the severity of accidents: A case study on freight-train derailments

Abstract

Severity assessment is considered as the key steps of risk assessment in process safety management. A procedure for the assessment and prediction of accident severity to improve industry safety is proposed in this paper. The procedure mainly includes the step that classifies and assesses the accident consequences based on casualties, losses, and emergency rescue (CLE) analysis, and establishes the prediction model to estimate the severity of accidents according to variables selected by the topology of activity-environment. The reliability validation of the procedure is performed using the mainly freight-train derailment data collected by the Federal Railroad Administration (FRA) from 2015 to 2023 as a case study. The results indicate that the severity of accidents can be quantitatively classified using the CLE method and scoring procedure. Fifteen variables were selected for the prediction model based on the established topology of freight-train derailment, and the accuracy of the best prediction model can reach 74%, confirming the good performance of the prediction model using RF machine learning. In addition, the application of the SHAP method enables the identification of critical variables that can credibly explain the contributions resulting in freight-train derailment severity findings and provide strategies for managers to minimize the severity of accidents.