A data-driven framework for railroad accident analysis based on the CRISP-DM and association rule mining: Empirical evidence from the federal railroad administration (2020–2024)

• 12,123 U.S. rail accidents (2020–2024) analyzed via CRISP-DM and Apriori ARM. • 705 rules with Lift > 1 grouped into Derailments, Crossings, and Other Impacts. • Derailments (51.9%) are mainly attributed to switch failure and human error. • Severe risks linked to hot, clear weather with or without active signalization. • High-risk months: June, September; critical states: Florida, Missouri, and California. The railway industry is a critical component of global transportation, yet it faces persistent safety challenges resulting in significant human and financial costs. While previous studies have investigated individual accident factors, the complex interactions among human, environmental, and organizational elements remain underexplored. To address this gap, the present study adopts a comprehensive, data-driven approach using the Cross-Industries Standard Processing for Data Mining (CRISP-DM) methodology to analyze 12,123 accident records from the United States Federal Railroad Administration (U.S. FRA) between 2020 and 2024. Descriptive trend analysis identified 2022 as the year with the highest human casualties and 2023 with the greatest financial losses. The Apriori algorithm was employed to extract 912 valid association rules. A Pareto-based prioritization yielded 705 high-lift and high-confidence rules, which were further refined through semantic pruning and validated via a 10-fold subsampling test. The rules were categorized into three main accident types: Derailments, Highway-Rail Crossings, and Other Impacts. Key findings reveal strong associations between severe accidents and conditions such as high temperatures, low visibility, and lack of signalization—particularly at highway-rail crossings. Some rule combinations demonstrated Lift values exceeding 3.2 and Confidence scores of 1.0, underscoring consistent, high-risk patterns. Spatiotemporal heatmaps further identified critical hotspots in Florida, Missouri, and California, with elevated accident rates in June and September. Visual analyses through scatter plots, heatmaps, and treemaps supported the extraction of interpretable, actionable insights. The study emphasizes the need for integrated, multi-factor safety interventions and provides a foundation for data-driven policymaking aimed at reducing railway accidents and improving systemic resilience in transportation networks . The Data-Driven Framework for Railway Accident Analysis