A HETEROGENEOUS DATA FUSION FRAMEWORK FOR TRAFFIC AND INCIDENT PREDICTION

Abstract

Presenting DataFITS, an open-source system for integrating traffic data from several sources, this article lays the groundwork for a complete dataset. To improve the performance and dependability of ITS systems, we propose a heterogeneous data fusion framework to increase the coverage and quality of information used in traffic models. Two applications that put our traffic estimate and incident categorisation methods to the test confirmed our predictions. Through a nine-month process, DataFITS fused four data types from seven sources in a spatiotemporal domain. Incident classification made use of the k-nearest neighbours (k-NN) method using Dynamic Time Warping (DTW) and the Wasserstein metric as distance measurements, whilst traffic estimation models used descriptive statistics and polynomial regression. The findings show that DataFITS enhanced the quality of information for up to 40% of all roads via data fusion and considerably expanded the coverage of roads by 137%. Using a polynomial regression model, traffic estimate reached an R2 score of 0.91. Incident classification, meanwhile, successfully classified three categories of events (accident, congestion, and non-incident) with an accuracy of about 80% and a 90% success rate on binary jobs (incident or non-incident).