On-device Malware Detection using Performance-Aware and Robust Collaborative Learning

The proliferation of the Internet-of-Things (IoT) devices has facilitated smart connectivity and enhanced computational capabilities. Lack of proper security protocols in such devices makes them vulnerable to cyber threats, especially malware attacks. Given the diversity and sophistication in malware samples, detecting them using traditional vendor database-based signature matching techniques is inefficient. This paper presents a collaborative machine learning (ML)-based malware detection framework. We introduce a) performance-aware precision-scaled federated learning (FL) to minimize the communication overheads with minimal device-level computations; and (2) a Robust and Active Protection with Intelligent Defense strategy against malicious activity (RAPID) at the device and network-level due to malware and other cyber-attacks. Deploying FL facilitates detecting malware attacks through collaborative learning and prevents data sharing, thus ensuring data security and privacy. RAPID denies the illegitimate user and aids in developing an effective collaborative malware detection model. A comprehensive analysis, results, and performance of the proposed technique are presented along with the communication overheads. An average accuracy of 94% is obtained with the proposed technique with 15% communication overhead, indicating 19% better performance than state-of-the-art techniques. Furthermore, the minimum accuracy drop of a model trained using RAPID is only 3% when 10% of devices are adversarial and 16% even when 40% of devices are adversarial.