Real time contextual collective anomaly detection over multiple data streams

Anomaly detection has always been a critical and challenging problem in many application areas such as industry, healthcare, environment and finance. This problem becomes more di cult in the Big Data era as the data scale increases dramatically and the type of anomalies gets more complicated. In time sensitive applications like real time monitoring, data are often fed in streams and anomalies are required to be identified online across multiple streams with a short time delay. The new data characteristics and analysis requirements make existing solutions no longer suitable. In this paper, we propose a framework to discover a new type of anomaly called contextual collective anomaly over a collection of data streams in real time. A primary advantage of this solution is that it can be seamlessly integrated with real time monitoring systems to timely and accurately identify the anomalies. Also, the proposed framework is designed in a way with a low computational intensity, and is able to handle large scale data streams. To demonstrate the e↵ectiveness and e ciency of our framework, we empirically validate it on two real world applications.