Learning from major accidents: A machine learning approach

This paper trains machine‑learning classifiers on historical major-accident databases (e.g., MHIDAS) to predict accident severity categories such as fatalities and injuries, comparing model families (Wide, Deep Neural Network, Wide&Deep). For PSM, it operationalizes “learning from major accidents” by turning investigation-record features into predictive decision support. The approach can help prioritize prevention/mitigation investments, improve consequence expectations during hazard reviews, and support triage during investigations.