Classification of imbalanced travel mode choice dataset with SMOTE and prediction using interpretable machine learning

Travel mode choices (TMCs) are dominantly investigated by utilizing traditional methods. TMC datasets are often imbalanced, with some modes significantly underrepresented, which leads to biased outcomes and poor performance on minority classes. This imbalance poses a significant challenge to both traditional and machine learning (ML) techniques. Therefore, the current study used the Synthetic Minority Oversampling Technique (SMOTE) to balance the TMC dataset and used several ML algorithms such as random forest (RF), decision tree (DT), Extreme Gradient Boosting (XGBoost), K-nearest Neighbors (KNN), and logistic regression for the prediction of TMC. Nonetheless, sensitivity analyses were performed to investigate the most significant feature for TMC prediction. The results show that applying SMOTE with ML algorithms, particularly RF with an accuracy of 0.937 and XGBoost, yielded superior performance, supporting SMOTE as an effective technique for improving classification accuracy in imbalanced datasets. However, logistic regression was found to be the least accurate among the models. Based on feature importance, it was found that features like driver’s license status, vehicle ownership, income levels, and urban/rural classification play significant roles in determining TMC. The current study provides both a methodological approach and actionable insights for improving TMC modeling, decision-making, and promoting a sustainable transportation system.