Traffic Jams Detection Using Flock Mining

The widespread use of GPS devices on cars enables the collection of time-dependent positions of vehicles and, hence, of their movements on the road network. It is possible to analyze such huge collection of data to look for critical situation on the traffic flow. The offline analysis of traffic congestions represents a challenging task for urban mobility managers. This kind of analysis can be used by the traffic planner to predict future areas of traffic congestions, or to improve the accessibility to specific attraction points in a city. Many traffic systems adopt ad-hoc sensors like cameras, induction loops, magnetic sensors to monitor the status of the traffic flows: these systems are very expensive for installation and maintenance, and they are restricted to the local monitoring of the road arcs where they are installed. On the contrary, the use of GPS data to check the traffic conditions requires low installation costs (a part for the installation on the vehicle) and it enables to virtually monitoring the entire road network. In this demo we present an innovative tool that exploits the data collected from GPSenabled cars to detect the occurrences of traffic jams on the road network. The detection of potential traffic jams is based on the discovery of slowly moving flock patterns, i.e. a set of objects slowly moving together for a minimum amount of time [2,1,5]. The tool has been integrated in the M-Atlas system [4,7] exploiting the implementation of the T-Flock algorithm provided by the system. To the best of our knowledge this is the first system that uses GPS data, combined with flock mining, to detect traffic congestions. Most of the approaches available in the literature for traffic analysis are based on aggregation of spatial or temporal data focusing on predefined areas [3]. It is important to point out that this tool does not provide real time analysis, but instead it allows the analysis of the historical data. We will sketch here a case study on a dataset of around 40,000 GPS-tracked cars in the surrounding of Pisa in Italy. The demo will highlight the tight integration of the spatio-temporal and data mining tools of the MAtlas system and the graphical user interface that assists the DM analyst in driving his/her analysis.