Traffic congestion forecasting is an important aspect of smart city governance and infrastructure development. The project aims at forecasting traffic volume by labeling it as low, medium, and high using the Metro Interstate Traffic Volume dataset. An end-to-end pipeline was created with data preprocessing, extensive feature engineering, and target processing with class balancing methods such as SMOTE. Different machine learning algorithms, such as Logistic Regression, Random Forest, XGBoost, LightGBM, and CatBoost, as well as deep learning algorithms like CNN, LSTM, GRU, and Bidirectional LSTM with Attention, were used. Stacking and Voting Classifiers ensemble techniques were also used to further improve prediction accuracy. Feature engineering methods, such as cyclical encoding, interaction features, lag variables, and rolling averages, greatly improved model performance. The XGBoost model was the best-performing with the highest accuracy at 93.27%, followed closely by the Stacking Classifier ensemble performing just slightly better at 93.41%. This research proves that the combination of sophisticated feature engineering, deep learning structures, and ensemble methods can significantly enhance traffic volume forecasting for effective traffic management, congestion control, and emergency response optimization.