Deep Learning Model Using Long Short- Term Memory (LSTM) for Time-Series Forecasting of Traffic Volume in Malaysia

Abstract

Malaysia has been facing traffic congestion problem for decades. The citizens waste about 250 million hours a year stuck in traffic. Extensive work has been conducted to address this issue in terms of identifying the possible remedial action based on predicted traffic outcome from a black-box model such as the Deep Learning (DL) model. However, most of them do not explain qualitatively the behaviour of a model and the prediction in a way that human can understand. They approximate the behaviour of a black-box by extracting relationships between feature values and the predictions using perturbation-based and backpropagation-based methods. The limitation on both methods, there is no qualitative mapping from the domain data that point the user to the prominent features set for a certain outcome. To address this limitation, the main objective of this project is to develop a new qualitative explainable DL model in predicting traffic volume in Malaysia. The results from this study can be beneficial to Malaysia Traffic Transportation Authority such as PLUS Malaysia Berhad to support the prediction and decision of traffic management in Malaysia. The newly proposed technique can also benefit the transportation authority to establish an Intelligent Transportation System (ITS) equipped with effective and explainable traffic prediction model.