Traffic congestion remains one of the problems that continue to arise, especially in urban areas, one of which is Bandung City, when the causes of the problem are not managed properly. Continuous management of the causes of congestion problems will result in a controlled traffic system for the foreseeable future. This condition can be achieved if there is a congestion classification prediction system available. A reliable prediction and classification system can support the government in formulating data-based traffic management strategies. The Random Forest and K-Nearest Neighbour machine learning classification methods are strengthened with time-based feature expansion to capture traffic behavior in various time frames, so that the objectives can be achieved. The dataset obtained from Area Traffic Control System Bandung includes traffic flow recorded at 15-minute intervals at several intersections. Additional features such as red light duration, road width, and spatial proximity to residential and commercial areas are included to improve model performance. The results show that the Random Forest classifier with time-based feature expansion outperforms K-Nearest Neighbours, achieving the highest performance of 96%. These results show the potential contribution in short-term traffic prediction and its effectiveness in supporting urban traffic planning and congestion mitigation efforts in Bandung.