ANOMALY DETECTION IN WIRELESS DEVICES USING CHANGE POINT ANALYSIS

To give end consumers a trustworthy experience, smartphones and IoT (Internet of Things) devices now need to have three key qualities: security, reliability, and availability. These qualities can be diminished by unrelated incidents or strange conduct that results in hardware damage, software changes, the theft of user information, and a negative influence on the speed or availability of the device. In light of these realities, this project focuses on exploiting smartphone power usage signals for anomaly identification. These signals indicate the device’s dynamic behaviour as a result of the interaction of various hardware elements that are simultaneously controlled by one or more applications. Due to the variations in its statistical features over time, this behaviour can be viewed as a non-stationary process. In light of this presumption, our methodology applies a changepoint detection theory-based feature extraction strategy. Then, it integrates three machine learning classifiers to maximise detection performance and add diversity. A dataset of simulated malware that was running in the background on a smartphone was used to validate the methods. AdaBoost, XGBoost, decision trees, and support classifiers have all been employed. The correctness of the provided data set is examined using these four key algorithms.