using random undersampling boosting classifier to estimate mode shift response to bus local network expansion and bus rapid transit services

This study proposed a machine learning-based classification method to accurately predict mode choice in response to potential strategies for transit promotion in a sprawling region. the method consists of a machine learning classifier, a genetic feature selection process, and statistical analysis process. the random undersampling boosting algorithm is adopted for imbalanced datasets in sampling. the genetic algorithm is applied to optimize the combination of independent variables grounded on the principle of maximum relevance and minimum redundancy. the 2017 national household travel surveys and the add-on samples data for the houston metropolitan statistical area in texas, usa, were utilized to build the mode choice classifier, which shows 99.22% classification accuracy for auto mode and 98.90% for transit mode. based on a comprehensive study of commuters’ trip characteristics and socio-demographics of the study region, bus transit network expansion and bus rapid transit strategies were proposed to stimulate the predominant single occupancy vehicle mode to be shifted to public transit. results show that the bus rapid transit, providing higher trip speeds for medium- and long-distance commuters, can significantly increase transit mode share by 8.24% and 8.95%, respectively. when the bus rapid transit is available to all the medium- and long-distance commuters, the total mode shift can increase to 15.96% in the study region. the walking distance to the nearest transit access is linearly associated with the mode shift to transit; up to 2.4% of current auto trips shifted to transit mode for those within a 5-min walking distance in the urban area.