Detection of Anomalous Behavior of Smartphone Devices using Change Point Analysis & Machine Learning

Security, reliability, and availability have become three fundamental characteristics that smartphones and IoT (Internet of Things) devices have to possess to provide end-users a trustworthy experience. These properties can be degraded by extraneous events or anomalous behavior provoking damage in hardware, changes in software, theft of user information, and impact of device performance in terms of speed or availability. Considering these facts, this paper focuses on anomaly detection on smartphones using their power consumption signals. These signals represent the dynamic behavior of the device due to the action of different hardware components controlled by one or many applications at the same time. This behavior can be seen as a non-stationary process due to the changes in time of its statistical properties. Considering this assumption, our methodology uses a feature extraction technique that is based on changepoint detection theory. Then, it fits three machine learning classifiers to inject diversity and maximize the detection performance. The methodology was validated on a dataset of an emulated malware running in the background of a smartphone. Our results have been compared with several power signal based approaches demonstrating that the proposed technique performs better in terms of detection accuracy.

Keywords: Anomalous Behaviour, Decision tree, XGBoosting and AdaBoost Classifiers  and Support Vector Classifier.