A Comparative Study of Various Traffic Flow Prediction Techniques Using ML Models and Real-Time Analysis

Due to the rapid growth of vehicles and the expanding urban landscape, the challenge of traffic congestion is becoming increasingly prominent. There has been a considerable shift towards data-driven solutions because of the availability of massive amounts of transportation-related data and the development of machine learning algorithms. In the proposed work, several feature sets for traffic flow are compared by using different ML methods: kNN, SVR, Decision tree, and Random Forest. The dataset used for the prediction is the traffic flow of the area between two hospitals in San Francisco. According to the result, feature extraction is more important than method selection to get accurate predictions. Selecting high quality features leads to fewer complex techniques which are easier to handle, are trustworthy and faster. Traffic flow prediction’s accuracy obtained by the decision tree surpasses the other methods. In addition to that, real-time analysis is applied to live feed captured with the help of CCTV. The live feed is then processed using the OpenCV library, the objects are detected using the YOLOv4 model which is further passed to DEEPSORT for vehicle tracking. This can further be processed by applying mathematical formulae to calculate the congestion on the road.