A Power Signal Based Dynamic Approach to Detecting Anomalous Behavior in Wireless Devices
Robin Joe Prabhahar Soundar Raja James, Abdurhman Ali Albasir, Kshirasagar Naik, Marzia Zaman, Nishith Goel · 2018
The health and security of wireless devices are fast gaining importance, and these are vital for effective implementation of sensor networks and Internet of Things (IoT). Any device, wired or wireless, needs a power source, and the power consumed is a consequence of its usage and functionality. In this context, this paper proposes a methodology to detect anomalous behavior of wireless devices by monitoring their power consumption patterns. The proposed methodology utilizes Independent Component Analysis (ICA) to extract information from the current power consumption of the device and generates features of the state of the device by calculating the degree of similarity of the extracted information with the known normal behavior of the device. Then, Recursive Feature Elimination (RFE) is used to select features from the generated feature vector. Finally, Classification algorithms are used to classify and detect the anomalous behavior. We have validated the methodology by emulating anomalous behavior on smartphones through a custom designed app that runs in the background while the main app is being used. Validation results indicate that the proposed methodology can be used to identify even a sparsely active malware existence with very high accuracy. The proposed model has an accuracy of 88% for a malware active for 1% of the total time and accuracy of almost 100% for malware active for 12% of the time.