Anomaly detection in data streams based on graph coloring density coefficients

Anomaly or outlier detection is a fundamental task of data mining and widely used in various application domains. The main aim of anomaly detection is to identify all the data points with significant deviation from other normal data points. Mining the outliers become more challenging in environments where data is received at extreme pace. Such environments demand detection of outliers on-the-fly mode. The existing anomaly detection methods focus more towards the identification of point anomalies, very few of them explore the problem of collective anomaly detection over data streams. In this paper, we present a mutual density based anomaly detection algorithm, which can identify collective anomalies in data streams. The concept of mutual density is adopted from the area of graph theory.