Identifying Risk Factors and Improving Rail Safety by Using Logistic Re gression and Ensemble Learning Approaches
Abstract:
Train derailments pose significant risks to safety, infrastructure, and the environment. Understanding the factors that contribute to these incidents is crucial for developing effective prevention and mitigation strategies. This study employs logistic regression to investigate the predictors of train derailments using the accident dataset from the Federal Railroad Administration (FRA). The model identified track type and presence of engineers as significant factors influencing derailment risk. Yard tracks, industry tracks and sidings were found to have higher odds of derailments compared to main tracks, emphasizing the need for targeted safety measures in these areas. Also, the presence of engineers was associated with reduced derailment odds, highlighting the importance of skilled crew in ensuring safe operations. This study also employs adaptive boosting, an ensemble learning technique to predict derailment accidents. The model accurately predicts 72% of all instances of derailment and non-derailment accidents. The learning model also identifies the gross tonnage of the train as a key factor in predicting the likelihood of the train derailing. These findings provide valuable insights for developing evidence-based interventions by railroad authorities and safety agencies to mitigate derailment risks and enhance railway safety.
