<p>Traffic congestion remains a significant challenge in large cities, exacerbated by the lack of effective prediction techniques and high-quality data. This study introduces HyTraNet, a deep learning model designed for traffic congestion prediction in road networks. The proposed model leverages a deep convolutional autoencoder-decoder architecture with Convolutional Long Short-Term Memory (ConvLSTM) layers to capture both spatial and temporal traffic patterns. Traffic data for selected locations in Delhi were collected as high-resolution images from Google Maps and preprocessed to enhance data quality. HyTraNet effectively extracts and analyzes congestion patterns from these images, providing accurate traffic forecasts. Experimental results demonstrate that the proposed model outperforms existing approaches in both computational efficiency and predictive accuracy, highlighting its potential for real-world traffic management applications.</p>