DTC: A framework to Detect Traffic Congestion by mining versatile GPS data

With increase in availability of GPS enabled devices, a large amount of GPS data is being collected over time. The mining of this data is likely to help in detection of the locations which face frequent traffic congestion. The prior knowledge of such locations will help the users in deciding whether or not to opt for that route. Avoidance of plying on such routes will also help in reducing the congestion in such locations. However, the authors feel that the work done so far in this field does not give very accurate results. The reason behind this is the inability of the work done so far to distinguish between jams and random short-term stoppages like traffic lights. To incorporate such differentiation in this paper, the authors propose an improvised traffic-jam-detection framework - DTC (Detect Traffic Congestion). This framework can be applied to versatile GPS data i.e. data coming from various kinds of devices like mobile phones, tablets or from vehicles etc. In the technique associated with this framework, these GPS data is first clusterized using the Expectation Maximization Algorithm. The clusters hence obtained are filtered out to acquire on-the-road vehicle data clusters. On further processing these clusters, a final binary output of either Traffic jam or Traffic light is obtained. The output is then fed to a J48 Classification Model to train it and hence make the predictions more accurate. The results obtained in the experiments are then cross-checked with the real-time data giving an accuracy of 86%.