Supervised learning data preprocessing for short-term traffic flow prediction

Abstract

The goal of the investigation is to develop and test a system capable of providing short-term (less than an hour) traffic flow predictions in an urban environment. We present a data acquisition and preprocessing pipeline capable of filtering and normalizing data collected using Here Maps API. The data is used for training a supervised machine learning model which is afterwards validated by observing actual road traffic conditions and making empirical observations on the predicted routes. All the experimental determinations were carried on the city of Timisoara, Romania. The traffic flow data collected and used is available as an open dataset.