A scalable distributed stream mining system for highway traffic data

To achieve the concept of smart roads, intelligent sensors are being placed on the roadways to collect real-time traffic streams. Traditional method is not a real-time response, and incurs high communication and storage costs. Existing distributed stream mining algorithms do not consider the resource limitation on the lightweight devices such as sensors. In this paper, we propose a distributed traffic stream mining system. The central server performs various data mining tasks only in the training and updating stage and sends the interesting patterns to the sensors. The sensors monitor and predict the coming traffic or raise alarms independently by comparing with the patterns observed in the historical streams. The sensors provide real-time response with less wireless communication and small resource requirement, and the computation burden on the central server is reduced. We evaluate our system on the real highway traffic streams in the GCM Transportation Corridor in Chicagoland.