Travel Mode Choice Prediction Using Imbalanced Machine Learning

Travel mode choice prediction is critical for travel demand prediction, which influences transport resource allocation and policies. modes are often characterised by severe class imbalance inequality, leads to the inferior predictive performance of minority bias in prediction. In existing studies, has not been addressed with a general approach. Basic resampling methods were adopted without much investigation, was assessed commonly used metrics (e.g., accuracy), suitable predicting highly imbalanced modes. To this end, paper proposes an evaluation framework systematically investigate combination six over/undersampling techniques three methods. case study using London Passenger Mode Choice dataset, results show that applying on substantially improves F1 score (i.e., harmonic mean precision recall) classes, considerably downgrading overall or model interpretation. These findings suggest combining statistical/machine-learning appropriate mode, effectively mitigates influence while achieving high accuracy addition, enriches options choice, would better support planning.