Intelligent prediction of traffic conditions via integrated data-driven crowdsourcing and learning

Abstract

Intelligent Transportation Systems (ITS) are gaining popularity among governments, businesses, and individuals due to their potential to make travel safer and more efficient. Machine learning for traffic prediction has emerged as a promising subfield of ITS, with the potential to aid in routing planning, congestion management, and urban development. Traffic infrastructure and mobile devices collect large amounts of heterogeneous data that can be used to predict traffic conditions, including real-time traffic data such as traffic camera images, speed measurements, and volume counts, as well as long-term static data such as speed limits, road conditions, and surrounding geography and infrastructure. Despite the availability of traffic data, many current machine learning models struggle to handle the wide variety of data types and to address both temporal aspects of real-time data and spatial aspects of long-term static data. To address this, we propose a new enrichment learning model that integrates dynamic data containing varying numbers of instances with static data to create an enriched fixed-length vector which can be used with other machine learning methods to improve performance and identify regions important for prediction. Results show that this novel enrichment learning model can improve the performance of traditional machine learning methods in the task of predicting future traffic speeds.