General Transit Feed Specification Assisted Effective Traffic Congestion Prediction Using Decision Trees and Recurrent Neural Networks

Abstract

Traffic congestion prediction is an open-ended critical problem highlighted by the rapid growth in intra-city transit mobility in recent years fuelling the necessity for an intelligent traffic management system in metropolitan areas. The majority of this research continues to rely on data from electronic devices and mobile signals, which can sometimes be manipulated to mislead the public. Modern state-of-the-art models of predictive analysis of Graph Neural Networks (GNNs), AutoRegressive Integrated Moving Average (ARIMA) and other Hybrid Deep Neural Networks have yielded positive results. However, determining which artificial intelligence model would be able to best address the issue of traffic congestion in densely populated areas. Based on this premise, we focus on using GTFS(General Transit Feed Specification) data and have constructed a meticulous and reflective dataset. We also postulate a study of multifarious models in comparison as well as a novel approach that maps traffic congestion as a classification problem rather than a regression-prediction problem to address the shortcomings of the issue. The highest accuracy metric for the optimised models was using the Decision Tree Classifier which yielded an accuracy of 81%. In this research article, we offer an overview of predicting traffic congestion whilst focusing on the G TFS dataset.