Propose adjustable Support Vector Machine approach for classifying imbalanced work travel mode choice data

• Investigation of travel mode choice (TMC) to work using conventional and modern techniques. • Proposed adjustable Support Vector Machine (SVM) for work TMC prediction. • Modern techniques outperformed the conventional methods where NN models predict a higher accuracy of 92%. • The SVM AK model outperformed the other ML models and enhanced the accuracy of typical SVM from 90.2% to 99.8% Investigating travel mode choice (TMC), particularly for work-related travel and its role in promoting sustainable transportation systems, is crucial in urban planning and policymaking to predict travel demands. Traditionally, the TMC data are imbalanced, and handling or learning from such data is usually challenging. Conventional techniques, such as the Multinomial Logit Model (MNL), are based on assumptions and are used for binary regression and classification. Several machine learning (ML) algorithms are based on binary classification; however, applying the theory of adjusting kernel scaling makes them capable of handling multi-class regression analysis. Therefore, the current study proposes an adjustable kernel (SVM AK ) based on adjusting kernel scaling theory and Support Vector Machine (SVM) to deal with the imbalance mode choice data. Besides, several ML algorithms such as standard SVM, Naive Bayes (NB), Gradient Boost (GB), Random Forest (RF), and Neural Network (NN) have been developed to compare the performance with SVM AK . The 2022 National Household Travel Survey (NHTS) – California data were used for the statistical analysis and empirical assessment. It was concluded that the modern techniques outperformed the conventional methods, where the SVM AK model performed better than the other ML models. Among all ML models, NN models achieve an accuracy of 92%, while SVM AK improves the typical SVM accuracy from 90.2% to 99.8% which shows the capability of handling an imbalanced TMC dataset. This approach can be used for supplementary transport planning tasks where there is an unequal distribution of data.