Traffic Congestion Prediction Using Multi-Layer Perceptrons And Long Short-Term Memory

Abstract

Traffic road congestion prediction is one of the solutions for solving the high rate of traffic congestion in many parts of the world. Many research works have been done to predict the traffic congestion, however those works use different data, context, and methods. This work explored possibility to use the CCTV footage to perform traffic prediction. The footage is processed automatically using object detection and object tracking algorithm to obtain traffic data. After that, the traffic data is modeled using both Multi-layer Perceptron (MLP) and Long Short-term Memory (LSTM). Model performance is measured by using Root Mean Squared Error (RMSE) to get the best approximation of the data. This study prove that automatically processed CCTV footage is indeed a viable option for traffic congestion prediction. The best model achieved RMSE value of 1.88, using MLP method and amounts of cars, buses, and trucks as predicted variable.