Predicting the most efficient routes for drivers is an imperative part of traffic organization because it helps cut down on delays. Using the Performance Measurement System (PEMS) dataset, this research proposes a capsule-weighted attention architecture for efficient traffic route prediction. Including data on traffic volume, speed, and occupancy, real-time data from the state of California can be found in PEMS, a popular traffic dataset. The proposed system includes the following features, among others: preprocessing the PEMS data; creating the CapsNet model architecture; training the model; and using the trained model to predict the optimal route for a given route. The suggested method is tested on the PEMS dataset, and its results are compared to those of more conventional statistical models for route prediction using mean absolute error and root mean absolute error. When compared to earlier statistical models, the new method proved to be more accurate in its prediction of the optimal course of action. The RMSE and MAE are reduced from the baseline CAPSNET of 29.13 and 15.21, respectively, when the residual block from the traffic flow capsule layer is incorporated into the proposed model. The results indicate that attention-based CapsNet architecture can be an effective method for optimal route prediction in traffic by making use of real-time data, which in turn can enhance traffic management and cut down on driver commute times.