This study explores machine learning approaches to predict multi-facet faults affecting operations of a railway network. Our study has established machine learning models and then benchmarked their performance including Extreme Gradient Boosting, Random Forest, and SVM. Field datasets have been collected over 6 years in collaboration with train operating companies and the infrastructure manager for operational, infrastructural, and environmental features. The main emphasis is placed on the high-speed railway network (linking three airports) in Thailand. Our research ensures robust model development through data preprocessing, data transformation, and then hyperparameter tuning. Our new results reveal that Extreme Gradient Boosting's superior predictive capability is evident, which can be attributed to its effective handling of non-linearity, feature interactions. The results also highlight the significant possibilities of the machine learning in proactive maintenance and mitigating risks. Additionally, our study is the first to explore the complex intercorrelation among vast amounts of historical accident data, recognising the intricacy that arises from technical, human-related, and environmental elements. Although some existing machine learning methods show promising outcomes, they encounter challenges when it comes to generalising and categorising complex data. Our study demonstrates the necessity of implementing innovative techniques to reveal intricate relationships in the accident data. This new approach results in improved prediction accuracy and contributes to creating a railway system that is safer and more reliable.
Kongsap et al. (Fri,) studied this question.