Machine Learning-Based Passenger Flow Prediction for Urban Public Transportation: A Case Study of Bus Networks

Abstract

Accurate passenger flow prediction is critical for optimizing urban public transportation operations, improving service quality, and reducing passenger waiting times. This study explores the application of three machine learning models—Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and Support Vector Regression (SVR)—in predicting short-term passenger flow of urban bus networks. Using one-month (March 2023) operational data from 50 bus routes in Guangzhou, including passenger count, departure time, weather conditions, and holiday information, we evaluate the models’ performance through metrics such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE). Results show that the GBDT model outperforms the other two: it achieves an MAE of 4.21, RMSE of 5.83, and MAPE of 6.78%, which are 18.3% and 25.6% lower in MAE than RF and SVR, respectively. The findings provide practical insights for public transportation agencies to adjust vehicle scheduling and improve resource allocation efficiency.

Next article in issue