Machine Learning Based Malware Detection in Wireless Devices Using Power Footprints

摘要

The adoption of smart devices, smartphones and IOT (Internet of things) devices, has witnessed a tremendous growth because of the convenience that they bring to users. However, at the same time, it makes them the target of malware developers and security attacks. Therefore, in this work, we propose a machine learning based methodology to detect malicious activities in wireless devices using power consumption footprints. The methodology consists of two main steps. First, a set of effective features are extracted from power measurements using both frequency spectral estimation and principal component analysis. Second, malware detection is formulated as a binary classification problem. Three different classifier algorithms are evaluated: support vector machines (SVM), naive Bayes (NB), and decision trees (DT). The SVM classifier shows better performance in terms of classification accuracy. Six different malicious behaviours are considered in the evaluation of the methodology process. The proposed methodology has successfully detected hard to spot malicious behaviours with (96-99)% accuracy.