Fast Event Detection on Big Time Series

Big data is exploding to facilitate our humans living by embedding smart devices everywhere, collecting realtime data, learning the daily habits and making the machines smarter. In addition to great advance in distributed computing with petabyte data, fast and real-time reaction on streaming data, which is know as fast event detection(FED) or anomaly detection, obtain wide attention which has a wide application in online fraud monitoring. In this paper, inspired by the time series analysis technique, a new algorithm of event detection is proposed to detect anomalous event. The proposed algorithm extensively reduce computation complexity of event detection from exponential to polynomial, which implies acceleration of more than thousand time. Verifications on four data sets confirm our theoretical prediction and promises fruitful results in further applications.