Evaluating the Benefits of Model Retraining for Self-Aware Vehicle Traffic Forecasting

Traffic congestion continues to cause costs and losses in time for commuters and businesses alike. Traditional approaches to traffic prediction often overlook the dynamic nature of driver behavior and road networks, leading to accuracy degradation in forecasting models. This paper presents a novel approach towards self-aware vehicle traffic forecasting, aiming to address the challenges posed by evolving traffic patterns. Leveraging machine learning algorithms and a comprehensive dataset derived from live traffic information systems, we propose the idea of a self-aware vehicle traffic forecasting system capable of adapting to changes in the environment to maintain prediction accuracy. Our approach involves regularly rebuilding forecasting models using recent data, coupled with parameter optimization techniques. Through a case study, we demonstrate that this adaptive approach can significantly improve the accuracy of traffic forecasts, achieving up to a $2 \%$ increase in overall accuracy and up to $4 \%$ for individual weeks. Our findings show the importance of continuous model adaptation in traffic forecasting systems. The presented idea lays the foundation for more robust and accurate traffic prediction systems with the potential to enhance the quality and efficiency of intelligent transportation systems.