Evaluation of Applicability and Accuracy of Bus Travel Time Prediction in High and Low Frequency Bus Routes Using Tree-Based ML Techniques

Abstract

Prediction of bus travel time is a key component of an intelligent transportation system and has many benefits for both service users and providers. Although there is a rich literature on bus travel prediction, some limitations can still be observed. First, high-frequency and low-frequency bus routes have different characterizations in both operational and passenger behavior aspects. Therefore, it is highly expected that bus travel time prediction methods for different frequencies must have different outputs. Second, in the era of big data, applications of machine learning (ML) techniques in travel time prediction have significantly increased. However, there is no single ML model introduced in the literature that is the most accurate in predicting bus travel, especially with regard to bus service frequency. Consequently, the main objective of this study is to determine the most applicable route construction approach and most accurate tree-based ML technique for predicting bus travel time on high- and low-frequency bus routes. The following tree-based ML techniques were adopted in this study: chi-square automatic interaction detection (CHAID), random forest (RF), and gradient-boosted tree (GBT). According to the results, CHAID was selected as the most accurate model for predicting travel time on high-frequency routes, while GBT showed the best performance for low-frequency service. CHIAD analysis identified distance between stops and terminal departure behavior as the most significant factors of travel time on high-frequency routes. Moreover, we introduced the "key stop-based" route construction method for the first time, which is an accurate, reliable, and applicable method.

Keywords

• machine learning
• tree-based models
• high-frequency route
• low-frequency service
• travel time prediction