رايحه سادات موسوي

This research was conducted with the aim of estimating the severity of weather-related railway accidents an‎d extracting spatial risk maps in order to identify high-risk hotspots for operational planning an‎d infrastructure adaptation. The case study focuses on the United States, an‎d the study concentrates on the severity of freight train derailments under the influence of climate change; U.S. railway accident data over a 30-year period from 1993 to 2024, including accident occurrences an‎d operational characteristics, were utilized, an‎d meteorological indices were assigned from the nearest weather stations to each accident. Accident severity was defined using three independent outputs: (a) the number of fatalities, injuries, an‎d damage (discrete numerical), (b) monetary loss (continuous), an‎d (c) the number of derailed wagons (five-class classification). All analyses were conducted in Google Colab using Python 3.12.12 an‎d stan‎dard data-mining libraries. The methodological framework includes statistical explorations of temporal, spatial, an‎d causal patterns, extraction of five-year association rules to distinguish stable an‎d emerging rules, an‎d predictive modeling. In the prediction component, baseline models including MRA, RF, SVR, an‎d XGBoost were first implemented, followed by sequential an‎d attention-based architectures including DNN, LSTM, BiLSTM, an‎d Attention (ATTN). The results indicate that memory-based an‎d attention-based models significantly outperform linear, tree-based frameworks an‎d simple DNNs; risk maps derived from the outputs of LSTM (continuous) an‎d ATTN (categorical) as the best-performing models exhibit strong spatial overlap in southeastern Texas, the Gulf Coast, an‎d the central an‎d northern plains with climatic changes in the United States. Feature importance analysis an‎d association rules confirm the dominant roles of temperature, maximum temperature, day of year, season, an‎d wind; this synthesis assists planners in prioritizing the identified spatial hotspots an‎d focusing appropriate operational measures precisely within those areas.

تغييرات آب و هوايي , سوانح ريلي , خروج از خط , يادگيري ماشين

Climate change , Railway accidents , Derailment , Machine learning

Rayehe sadat Mousavi

Morteza Bagheri