A Heterogenous System for Traffic Prediction

Abstract

Recent developments in the field of Machine Learning have resulted in great progress, both academically and commercially, in various applications. The success of these applications comes from being able to either predict or categorize series or elements that, before, could not be analyzed using classical statistical methods, thus solving complex, real problems that people are facing daily. Traffic is one of the most common issues in urban areas. It causes delays and frustration to both people and businesses, impacting billions across the globe. It is one of the greatest problems of the contemporary world. We propose a heterogeneous, distributed system that is able to provide traffic predictions at scale, with tremendous precision and consistency, using novel learning models and cloud-based system technologies. In a case study on data collected from Puget Sound, Washington, in 2017, we measure the precision of our novel system at an average of below 1 km/h. At the 90th percentile, the model is still able to provide valuable predictions, with an absolute error under 5 km/h.