A neural network approach for traffic prediction and routing with missing data imputation for intelligent transportation system

Abstract

A robust traffic rerouting system is important in traffic management, alongside an accurate traffic simulation model. However, missing data continues to be a problem as it will inevitably cause errors in predicting the congestion levels, resulting in a less efficient rerouting. The lack of a realistic traffic simulation also serves to hamper the development of a better traffic management system. As such, this paper aims to address both problems by proposing three solutions: (i) a traffic simulation that would model a live-traffic, (ii) a pheromone-based, neural network traffic prediction and rerouting system, and (iii) a missing data handling method utilising weighted historical data method named Weighted Missing Data Imputation (WEMDI). The traffic simulation model was benchmarked using Google Maps rerouting system. WEMDI was tested by comparing the performance of the rerouting system with and without WEMDI’s integration for various levels of missing data. The results showed that the traffic simulation model displayed a high correlation to that of Google Maps, and the WEMDI-integrated system displayed 38% to 44% improvement in the related traffic factors, when compared to a situation with no rerouting system in place, and up to 19.39% increase in performance compared to the base rerouting system for missing data levels of 50%. The WEMDI system also displayed robustness in routing other locations, displaying a similarly high performance.