Time Prediction Model Based on NN ‐ SVM and H∞ Filter

Considering the limited prediction accuracy of the existing time prediction model, we propose a bus arrival time prediction model based on improved SVM and H∞ filter technique. The NN‐SVM which had two input features was used to predict the baseline of bus running time from historical trip data. Applying the real‐time running information and combining it with baseline time, we use H∞ filter to predict bus arrival time dynamically. Bus arrival time forecasted by the proposed model was assessed with the data of transit route number 142 in Changsha in China. Results demonstrate that the proposed method achieved a Mean Absolute Error (MAE) of 19.47 s, Root Mean Square Error (RMSE) of 23.42 s, representing a 65.0% reduction in MAE and a 74.5% reduction in RMSE compared to standard SVM. and PBIAS of −2.4%, indicating high accuracy with minimal underestimation bias. Compared with conventional SVM and Kalman‐based models, the proposed model reduced RMSE by 19.4% and improved robustness in dynamic traffic scenarios.