Neural Network-based Traffic Situation Classification for Intelligent Transportation Systems

The intelligent transportation systems (ITS) rely heavily on traffic state identification for congestion management, traffic control, and real-time decision-making. The increasing availability of traffic sensor data has enabled the use of data-driven learning frameworks to capture nonlinear traffic dynamics. This paper proposes a traffic classification system based on neural networks combining vehicle count and temporal features to classify traffic into low, normal, heavy, and high levels. The model incorporates feature engineering based on temporal attributes and is trained using a feedforward neural network with Adam optimization and cross-entropy loss. Experimental results demonstrate an accuracy of approximately 82%, with confusion matrix analysis confirming strong class discrimination. The framework is effective for real-time traffic monitoring and smart city applications.