Implementation of the Convolutional Neural Network (CNN) Method to Detect Traffic Congestion and Determine the Best Route Using the Heuristic Search Algorithm on Palembang City Highways

The city of Palembang is facing a significant increase in traffic density, causing various problems such as congestion. Research was conducted to overcome this problem by using various models and algorithms for object recognition, vehicle counting, traffic density prediction, and finding the best route. The YOLOv8 model was developed using 3592 images to recognize five object classes, with a training accuracy of 95% and testing accuracy of 93.83%. Using a combination of YOLOv8 and DeepSORT in calculating vehicles from 72 videos taken over 4 days shows an average accuracy of 95.51% for motorbikes and 79.04% for cars. Next, the CNN model was applied to classify road conditions with a training accuracy of 83.98% and testing accuracy of 89.22%. Traffic density prediction using CNN achieves 77.7% accuracy. To find the best route, the heuristic search algorithm (A-Star) is used, determining the optimal route from the Ampera bridge to Sultan Mahmud Badaruddin II airport based on existing road conditions. In conclusion, this combination of models and algorithms is effective in overcoming traffic problems in Palembang, showing satisfactory results in object recognition, vehicle counting, traffic density prediction, and determining the best route.