A Comparison of Machine Learning Models in Predicting Competition Between High-Speed Rail (HSR) and Air Transport: The Tehran-Mashhad Case Study

Mode choice modeling represents a fundamental domain within transportation engineering and urban-regional planning. The competition between high-speed rail (HSR) and air travel holds particular significance for demand forecasting, revenue optimization, environmental policy, and infrastructure development. Traditional discrete choice models have long served as the cornerstone of mode choice analysis. These models offer interpretability, compatibility with stated and revealed preference data, and the capacity to compute policy-relevant elasticities. However, they suffer from critical limitations, such as the independence of irrelevant alternatives (IIA) assumption and inability to accommodate large, noisy datasets. Conversely, Machine Learning (ML) methodologies have gained prominence for their capacity to handle complex, nonlinear, and high-dimensional data. By applying Artificial Neural Network (ANN), Random Forest (RF), Decision Tree (DT), and K-Nearest Neighbor (KNN), this study addresses two critical research gaps: (۱) the scarcity of ML applications in developing countries with limited HSR infrastructure, and (۲) the limited incorporation of psychological factors alongside socio-economic variables. Using stated preference data from ۱۰۰ Iranian respondents across ۱۸ travel scenarios, this research develops ML-based models, examining variables such as travel time, cost, income, service frequency, previous travel experiences, and psychological factors including fear of flying. The findings reveal that ANN emerged as the top performer with an overall accuracy of ۸۴.۶۷%. The RF model followed with ۸۲.۴۴% accuracy, showing robust predictive capability with relatively balanced class-wise performance, though slightly favoring the majority class. Also, class-specific analysis across all models consistently demonstrated higher precision for airplane predictions.

کلیدواژه ها:

High-Speed Rail ، Mode Choice ، Machine learning ، Classification Report