Keywords: train derailment, clustering, machine learning, freight train, k-nearest neighbors, support vector machines, random forest, gradient boosting.

This study aims to estimate freight train derailment severity using the U.S. FRA rail accident database spanning from 1997 to 2023. After preprocessing, which included data cleaning and normalization, the dataset comprised 3967 records. The data was split into training (80%) and testing (20%) sets. Using the NBclust function in the R programming environment, optimal clustering for causes was determined, resulting in four clusters based on specific criteria. Each cluster was analyzed using four machine learning techniques: K-Nearest Neighbors (KNN), Support Vector Machines (SVM), Random Forest, and Gradient Boosting. The results were aggregated based on cluster weights derived from the clustering process. Performance evaluation metrics, including RMSE and Accuracy, were used to assess the models. The findings indicate that all classifiers performed well, with KNN demonstrating superior performance, achieving an accuracy of 92.36% and an RMSE of 0.26. Additionally, the proposed model's average accuracy of 91.53% outperforms the previous benchmark study, which reported an average accuracy of 79.56%. These results suggest that the proposed model is effective in estimating derailment severity and can be a valuable tool for railway safety management.

download the full-text of this paper (PDF, 10 pages, 543 Kb)

go to the previous paper
return to the table of contents
return to the volume description