Traffic Prediction: A Comparison between the LSTM and Multi-Layer Perceptron Algorithm

Abstract

Every year, there is a rapid increase in population and vehicle purchases. The increasing number of vehicles also increases the traffic flow prediction rate on urban roads. One popular method is deep learning, a frequently used method for making traffic flow predictions in recent years. In this paper, We proposed a comparison between 2 popular algorithms, Long-Short Term Memory (LSTM) and Multi-layer Perceptron, to find the better algorithm for quick prediction. The input model used to make the forecasting model is the number of vehicles every hour. We calculate the predicted number of vehicles passing at a given time to perform the traffic condition prediction process. To see the performance of both models, train and test using a public dataset from Kaggle. We split the data into 80% for training and 20% for testing data. The dataset is divided into two columns, each with a timestamp of the date and time of data collection and containing the number of vehicles at a specific time with a 5-minute interval. The model trained these algorithms and achieved RMSE value from LSTM for 27.02 and multi-layer perceptron for 7.256. As a result, the multi-layer perceptron model produces significantly better predictions than the LSTM model.