Support vector Machine and back propagation neutral network approaches for trip mode prediction using mobile phone data

Abstract

This study provides a methodology to identify travellers’ transportation modes by tracking the mobile phone data, which aims to obtain the accurate mode split rate for providing decision support in urban traffic planning. First, the effective mobile phone singling data and GPS data are collected from the communication operators and a mobile phone app, respectively. Considering the differences in velocity and acceleration of different trip modes, a trip mode characteristic description model is built based on wave characteristics and moving average method. Compared with the wave characteristics, the moving average method shows a better accuracy of 90%. Then training samples are drawn by two data selection methods including probability proportional to size sampling and equal amount sampling. Furthermore, the classifier method for mode choice prediction is developed by support vector Machines (SVMs) and back propagation neutral network. Finally, the results of the case study show that using a 30-point moving average training data set can improve the prediction accuracy largely, and the SVM method gets a better accuracy of 82%. The potential of using the mobile phone data to build a new mode choice prediction method in the field of transportation is shown.