Predictive Modeling of Vehicle Speed Using Machine Learning: A Case Study with Traffic Flow Features

Vehicle speed is the most representative parameter reflecting traffic congestion status, and predicting vehicle speed through relevant traffic flow state parameters is a hot research issue in the field of transportation. This paper adopts three machine learning models, namely decision tree, random forest and XGBoost, to carry out predictive modeling and experimental analysis on the vehicle speed of the expressway entrance road in Bao'an District, Shenzhen. The experimental data comes from the Shenzhen Municipal Government Data Open Platform. By comparing the performance indicators (MSE, RMSE and R²) of the three models, it is found that the XGBoost model performs the best and can better capture the complex nonlinear relationship between vehicle speed and multi-dimensional traffic features. In addition, SHAP values are used to analyze the impact of key features on vehicle speed. The results show that light vehicle traffic has a significant positive effect on vehicle speed, while lane occupancy rate shows a strong negative impact. The results show that the vehicle speed prediction method based on machine learning has high accuracy and practicality, providing decision-making basis for traffic management departments to formulate traffic optimization strategies.