Pattern mining from historical traffic big data

Abstract

Knowledge mining from the historical traffic big data is absolutely necessary for future intelligent transportation system (ITS) and smart city. Mining traffic data is a challenging task that can be used for traffic forecasting and improving traffic flow. In this paper, we explore and analyse the historical traffic big data to extract the informative patterns. Three years (2013 to 2015) real traffic data was collected from the city of Porto, Portugal. We developed a Java based traffic data observation (TDO) tool for visualising traffic data, which can filter and extract expressive patterns from the traffic big data based on input features. Then, graphs are generated by TDO from the traffic data to find the historical averages of traffic flow. Finally, we have applied regression models: Linear Regression, Sequential Minimal Optimisation (SMO) Regression, and M5 Base Regression Tree on the traffic data to find annual average daily traffic (AADT) and compare their results. Also, we have used regression trees to find the traffic patterns. The goal is to find the abnormal traffic patterns from the historical traffic big data and analyse them to improve the traffic management system (TMS).