Traffic Occupancy Prediction Using a Nonlinear Autoregressive Exogenous Neural Network

Abstract

The main aim of intelligent transportation systems is the ability to accurately predict traffic characteristics like traffic occupancy, speed, flow, and accident based on historic and real-time data collected by these systems in transportation networks. The main challenge of a huge quantity of traffic data collected automatically, stored, and processed by these systems is the way of handling and extracting the required traffic data to formulate the prediction traffic characteristic model. In this research, the required traffic data of a specified road link in the UK are extracted from the big raw data of the Split, Cycle, and Offset Optimization Technique (SCOOT) system by designing a C++ extractor program. In addition, short-term traffic prediction models are created by using a deep learning technique called a Nonlinear Autoregressive Exogenous (NARX) neural network to find accurate and exact traffic occupancy. Three scenarios of time intervals which are 10 minutes, 20 minutes, and 30 minutes are considered for analyzing the prediction accuracy. The results showed that the prediction models for the 30 minutes interval scenario have very good accuracy in estimating the future traffic occupancy compared to other scenarios of time intervals. In addition, the testing and validation study showed that the prediction models for 30 minutes intervals for particular road link yield better accuracy than 10 minutes and 20 minutes intervals.