Anomaly detection using the Kullback-Leibler divergence metric

Abstract

A method of detecting changes or anomalies in periodic information-carrying signals or any other sets of data using Kullback-Leibler divergence is described. Theoretical reasons for using this information-theoretic approach are briefly outlined and followed by its detailed application on disturbance/anomaly detection in wireless signals. Even though the concept is illustrated in a communications centric framework, it is more generally applicable in areas such as computational neuroscience, mathematical finance and others where it is important to statistically detect unexpected signal distortions. The results obtained show that the proposed approach is robust, highly effective, and has a low implementation complexity.