Traffic congestion remains a major challenge worldwide, affecting economies, the environment, and urban life. This study explores how machine learning can predict congestion by incorporating road user behavior. We generate synthetic traffic data simulating driving scenarios, capturing key factors like lane changes and speed. Our model combines CNNs and LSTMs with a dense layer for classification. The CNN extracts spatial features, while the LSTM captures temporal patterns. An attention mechanism focuses on important features. Trained on diverse scenarios, our model achieves 98% accuracy, outperforming traditional models and providing a behavior-aware approach to forecasting. Validation with synthetic data highlights the potential for improved traffic management. Our findings highlight how neural networks can enhance urban traffic analysis, providing insights into how road user behavior impacts congestion and laying the groundwork for future research on smarter traffic solutions.