شقايق نامداري جبارابادي

In most countries worldwide, railway transportation systems play a significant role in economic development. Although railways are currently the safest mode of ground transportation, railway accidents continue to occur, impacting transportation safety. Therefore, assessing the occurrence of railway accidents an‎d managing infrastructure to address potential risks is a valuable research topic. Today, with advancements in artificial intelligence an‎d big data, valuable information can be effectively extracted from historical accident data, which is crucial for understan‎ding an‎d preventing accidents. The sample size of accident data can vary from hundreds to thousan‎ds, depending on the time period an‎d region in question. The data analyzed an‎d examined in this study were obtained from the U.S. Federal Railroad Administration (FRA) accident database. This database records all incidents that cause specified damages to rail equipment, signals, track structure, an‎d roadbed. Each accident record includes information on 144 features that detail the accidentʹs conditions, such as location, time, weather conditions, track class, method of operation, train identification number, etc. Furthermore, the FRA dataset includes both structured data (e.g., numerical data) an‎d unstructured data (e.g., descriptive text data). Studies conducted on this dataset from January 1, 2009, to June 30, 2020, indicate that derailment accidents account for over 70% of the U.S. railway industryʹs average annual financial damages. Consequently, the ability to differentiate accident types an‎d rank features associated with accident causes can contribute to more economical risk management strategies. This study involves an examination of various ensemble learning methods an‎d their applications in predictive modeling. In the subsequent sections, the dataset is eva‎luated, an‎d issues such as data cleaning an‎d missing data imputation are addressed. A comparative analysis of the results from different ensemble models, particularly the AdaBoost-Bagging model, is then conducted. Furthermore, efforts to improve prediction accuracy through clustering an‎d model hyperparameter tuning were implemented.

پيش‌بيني سوانح ريلي , يادگيري تركيبي , يادگيري ماشين , ايمني راه‌آهن , داده‌كاوي

Railway accident prediction , Ensemble learning , Machine learning , Railway safety , Data mining

Shaghayegh Namdari

Dr. Masoud Yaghini