This research investigates the use of sophisticated machine-learning methods to forecast traffic patterns at various urban intersections. By leveraging a detailed dataset collected from strategically situated sensors, this study applies Gated Recurrent Units (GRUs), a type of Recurrent Neural Network (RNN) designed for sequence prediction, to anticipate traffic volumes. The data includes vehicle counts, environmental factors, and time markers from multiple junctions over an extended period. After thorough data preprocessing, feature engineering, and diligent model training, the research demonstrates how GRUs can effectively manage the temporal and sequential relationships present in traffic flow data. The predictive models were assessed using the Root Mean Square Error (RMSE) metric, which showed notable differences in predictive accuracy at different urban junctions. This study adds to the expanding knowledge in traffic management systems and offers a solid framework for real-time traffic forecasting, with the goal of improving urban mobility and alleviating congestion through smart traffic management solutions.