Elderly Travel Mode Choice in Thailand-Evaluating MNL and Machine Learning Models

Anantaya Philuek¹Panuwat Wisutwattanasak¹Fareeda Watcharamaisakul¹Chinnakrit Banyong¹Anon Chantaratang²Thanapong Champahom²Vatanavongs Ratanavaraha¹Sajjakaj Jomnonkwao^1*

• ¹Suranaree University of Technology, Nakhon Ratchasima, Thailand
• ²Rajamangala University of Technology Isan, Nakhon Ratchasima, Nakhon Ratchasima, Thailand

This investigation analyzes the determinants of transportation mode selection among elderly populations in Thailand through a comparative approach utilizing both traditional statistical modeling and contemporary machine learning techniques. The research compares the predictive effectiveness of the Multinomial Logistics Regression (MNL) model against advanced algorithmic approaches including Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Categorical Boosting (CatBoost) in forecasting elderly travel behavior. The research utilizes a dataset comprising 1,000 elderly participants distributed across Thailand's four geographical regions, with data collection conducted via structured questionnaires encompassing demographic factors, journey purposes, frequency of travel, expenditure considerations, and modal preferences. Results indicate that the Random Forest algorithms achieved the highest predictive performance on the comprehensive dataset (99.83% accuracy), while CatBoost demonstrated excellent performance on test data (94%). Shapley Additive Explanations (SHAP) analysis identified transportation expenditure, travel party size, temporal considerations, and economic status as the predominant variables influencing modal selection decisions, with lower-income elderly individuals showing a greater tendency for public transportation utilization relative to their more affluent counterparts. The MNL model revealed transportation cost as the most statistically significant predictor of public transit usage (p<0.001), indicating that elderly individuals confronted with elevated travel expenses tend to substitute public transportation with private vehicle alternatives. Concurrently, machine learning methodologies demonstrated enhanced capacity to capture complex relationships between predictive factors and exhibited superior predictive accuracy compared to conventional MNL modeling. These findings offer important implications for the formulation of age-sensitive public transportation policies, particularly emphasizing cost reduction strategies and infrastructure enhancements designed to accommodate the specific mobility requirements of elderly populations.Country Methodology Data Used Key Findings Accuracy (%) Wei et al. (2024) China Random Forest Bus ridership data