DATAFITS: LEVERAGING HETEROGENEOUS DATA FUSION FOR ACCURATE TRAFFIC AND INCIDENT FORECASTING

Abstract

In order to create a complete dataset, this study presents DataFITS (Data Fusion on Intelligent Transportation System), an open-source framework that gathers and combines traffic-related data from several sources. According to our hypothesis, a heterogeneous data fusion framework may improve the quality and breadth of information for traffic models, boosting the effectiveness and dependability of applications for Intelligent Transportation Systems (ITS). Two applications that made use of event categorisation and traffic estimate models confirmed our hypothesis. Over the course of nine months, DataFITS gathered four different kinds of data from seven sources and combined them in a spatiotemporal domain. While incident categorisation utilised the k-nearest neighbours (k-NN) method with Dynamic Time Warping (DTW) and Wasserstein metric as distance measurements, traffic estimation models used polynomial regression and descriptive statistics. According to the results, DataFITS enhanced information quality for up to 40% of all roads via data fusion and dramatically expanded road coverage by 137%. Using a polynomial regression model, traffic estimation obtained an R2 score of 0.91, while incident classification reached 90% accuracy on binary tasks (incident or non-incident) and about 80% accuracy on categorising three distinct event categories (accident, congestion, and non-incident).