Context-Aware Anomaly Detection in Embedded Systems

Abstract

To meet the reliability of embedded systems, fault-tolerant methods are widely used. The first step in many of these methods is detecting faults and anomaly detection is often the primary technique which leads to early indication of faults. In the context of embedded systems, some anomaly detection methods are available however; none of them are adaptable to dynamic environments. All of the previous works attempt to provide anomaly detection systems without considering the context of the data. Contextual anomalies, also referred to as conditional anomalies, have different behavior in different contexts. The purpose of designing a context-aware anomaly detection mechanism is to provide the capability of detecting anomalies while the system’s environment changes. In this paper, a method for detecting anomalies is proposed which adapts itself to the changes in dynamic environments during detection phase. This method first gives the context of a small window in a data flow and then loads corresponding configuration to the anomaly detector. The results have shown an average of 68.83% of true positive rate and 11.41% of false alarm rate.