Cluster-Based Logistic Regression Model for Holiday Travel Mode Choice

Abstract

With the rapid growth of holiday travel market, more and more attention has been paid to the analysis of holiday travel behavior, such as the holiday travel mode choice. This study presents a cluster-based logistic regression model for predicting the travel mode choice on holiday. At first, a regression and classification tree approach is employed to split the source date to clusters. Based on the data collected from the Beijing Fragrant Hills Park during the Qingming Festival (Tomb-Sweeping Day), an optimal tree with two levels and three leaf nodes is built and the collected data are divided into three clusters according to the tree structure. The three clusters are further considered as the dummy variables for the logistic regression analysis. Since the cluster based logistic regression model avoids the variable interaction effects, it significantly outperforms the logistic regression model in terms of prediction accuracy.