Application of Machine Learning Classifiers for Mode Choice Modeling for Movement-Challenged Persons

Abstract

In this study, we aimed to evaluate the performance of various machine learning (ML) classifiers to predict mode choice of movement-challenged persons (MCPs) based on data collected through a questionnaire survey of 384 respondents in Dhaka, Bangladesh. The mode choice set consisted of CNG-driven auto-rickshaw, bus, walking, motorized rickshaw, and non-motorized rickshaw, which was found as the most prominent mode used by MCPs. Age, sex, income, travel time, and supporting instrument (as an indicator of the level of disability) utilized by MCPs were explored as predictive variables. Results from the different split ratios with 10-fold cross-validation were compared to evaluate model outcomes. A split ratio of 60% demonstrates the optimum accuracy. It was found that Multi-nominal Logistic Regression (MNL), K-Nearest Neighbors (KNN), and Linear Discriminant Analysis (LDA) show higher accuracy for the split ratio of 60%. Overfitting of bus and walking as a travel mode was found as a source of classification error. Travel time was identified as the most important factor influencing the selection of walking, CNG, and rickshaw for MNL, KNN, and LDA. LDA and KNN depict the supporting instrument as a more important factor in mode choice than MNL. The selection of rickshaw as a mode follows a relatively normal probability distribution, while probability distribution is negatively skewed for the other three modes.

Keywords

• persons with disability
• mode choice
• movement-challenged person
• machine learning
• Dhaka