Akademik Veri Yönetim Sistemi
TÜBİTAK Projesi, 2219 - Yurt Dışı Doktora Sonrası Araştırma Burs Programı, 2022 - 2023
This study investigates the mode choice behavior of individuals undertaking non-work and non-school ("other-related") urban trips by employing advanced variable selection and machine learning techniques. Using household travel survey data from the 2015 Eskişehir Transportation Master Plan, mode choice models were developed and compared based on their predictive performance and interpretability. Eight modeling approaches were employed, including traditional logistic regression (LR), regularized regression techniques (Ridge, Lasso, Adaptive Lasso, and Elastic Net), stepwise regression, and non-parametric machine learning models (Random Forest and Support Vector Machine). The analysis focused on 43 explanatory variables representing individual, household, neighborhood, and land use characteristics. Multicollinearity diagnostics revealed that conventional logistic regression assumptions were not met, underscoring the necessity of regularization and variable selection methods. The results showed that Random Forest and SVM models achieved the highest predictive accuracy (ROC ≈ 0.89), while Lasso-based models provided a good balance between prediction power and interpretability. Key determinants of mode choice included job status, driver’s license ownership, vehicle availability, trip frequency, and household income, whereas variables such as trip distance and population density were found to be insignificant in the context of Eskişehir’s compact urban structure. The study highlights the value of interpretable machine learning models for understanding urban travel behavior and informs transportation planning strategies targeting flexible trip purposes beyond routine work and school commutes.