Logistic regression model for discretionary lane changing under congested traffic

Assessing the probability of lane-changing (LC) is essential to traffic simulation for more realistic representation of complicated traffic phenomena in congested traffic. Discretionary lane changes (DLC), which are not required for reaching a destination, are decided by drivers for the purpose of faster travel. The probability of DLC is related to the speed difference and density difference between adjacent lanes. To reveal the characteristics of DLC, we aggregated Next Generation Simulation trajectory data into a mesoscopic scale and analysed them to obtain detailed information on traffic conditions near DLC actions. We first constructed a joint probability distribution of LC in the speed difference and density difference domains. The distribution shows that speed difference and density difference significantly influence the LC rate. The constructed distribution indicates that drivers tend to change their lanes more frequently to travel faster and to have more space in the target lane. To quantify these relations, logistic regression was applied to the collected data. The outcomes of logistic regression show that the relations are statistically significant and non-linear.