The Travel Time Prediction by Machine Learning Methods with Traffic Data in Chiayi City, Taiwan

This study proposed three approaches to predict the travel time in urban corridors. Data used in this study include traffic data collected by vehicle detectors (VD), real-time bus operation data from onboard unit (GPS and OBD II) (eBus) and Cellular-based Vehicle Probe (CVP) data from telecom companies. Data were collected from October 1, 2018 to March 4, 2019 and divided into two groups for experiments: training data for building models and validation data to evaluate the performance of models. Three methodologies used in prediction are Gradient Boosting Regression Tree (GBRT), K-Nearest Neighbors (KNN) and Linear Regression (LR). The experiment results indicate all three methods perform well in the prediction of travel time for urban corridors. The quality of data also has impact on the prediction performance: among them, CVP data deliver the best prediction results.