Signal Processing-based Anomaly Detection Techniques: A Comparative Analysis
Résumé
In this paper, we present an analysis for anomaly detection by comparing two well known approaches, namely the Principal Component Analysis (PCA) based and the Kalman filtering based signal processing techniques. The PCA-based approach is coupled with a Karuhen-Loeve expansion (KL) to achieve higher improvement in the detection performance; on the other hand, based on a Kalman filter, we built a new method by combining statistical methods such as: gaussian mixture and a hidden markov modellers, which allows us to obtain performances better than those obtained with the PCA-KL expansion method. For this newer method, our approach consists of not assuming anymore that the Kalman innovation process is gaussian and white. In place, we are assuming that the real distribution of the process is a mixture of normal distributions and that, there is time dependency in the innovation that we will capture by using a Hidden Markov Model. We therefore derive a new decision process and we show that this approach results in an considerable decrease of false alarm rates. We validate the two comparative approaches over several different realistic traces.