ITSC 2025 Paper Abstract

Paper TH-LM-T30.6

Mohebbi Najmabad, Mohammadreza (Josef Ressel Center Vision2Move, University of Applied Sciences ), Davoudi Jaghargh, Niloofar (Ferdowsi University of Mashhad), Döller, Mario (University of Applied Sciences FH Kufstein Tirol), Hassan Zada, Mohammad Javad (Technical University of Munich), Yamnenko, Iuliia (Technical University of Munich, TUM School of Engineering and De)

Network-Wide Traffic Prediction Via Bayesian State-Space Neural Networks and ARIMAX

Scheduled for presentation during the Regular Session "S30a-Intelligent Modeling and Prediction of Traffic Dynamics" (TH-LM-T30), Thursday, November 20, 2025, 12:10−12:30, Gold Coast

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Keywords AI, Machine Learning for Real-time Traffic Flow Prediction and Management, Data Analytics and Real-time Decision Making for Autonomous Traffic Management, AI, Machine Learning Techniques for Traffic Demand Forecasting

Abstract

Accurate prediction of critical traffic variables, such as flow, speed, and occupancy, is crucial to maximize efficient urban traffic management and enable Intelligent Transportation Systems (ITS). However, traffic data are often high-dimensional and subject to spatio-temporal dependencies, complicating accurate predictions. In this paper, a hybrid framework is presented in which a Bayesian State-Space Neural Network (BSSNN) is combined with ARIMAX to improve the forecasting of traffic variables. BSSNN effectively captures nonlinear interactions in the data, while ARIMAX refines the prediction by capturing the structured residual dependencies. The experiment proves that the designed hybrid framework outperforms traditional machine learning and deep learning models considerably by providing higher prediction accuracy. The performance of the model was assessed on the metrics of Mean Squared Error (MSE), Mean Absolute Error (MAE), and coefficient of determination (R2), with the latter having a high value of 0.965. The high value R2 proves the stability and therefore the generalizability of the proposed model across varying prediction horizons. The results indicate that this proposed hybrid framework is highly compatible with actual ITS applications and can be extended to use external input data along with adaptive learning protocols.