Publication

Abstract : Sensors in the smartphone play a vital role in various user-friendly mobile services. The mobile application requires user permission to access the permission imposed sensors and not for other sensors. The sensors in the smartphone are vulnerable to various attacks. The attackers can exploit these sensors to trigger malware, extract the sensitive information of users and other nearby devices, and expose users’ confidential information. We propose SBTDDL, a novel context-aware framework for detecting sensor-based threats on Android smartphones using deep learning. In our work, (a) we identify the sensor-based threats by using the state (on and off) of the sensors in the smartphone for different user activities, (b) binary classification is performed in the sequence prediction model to classify the benign and malicious activities on the device, (c) SBTDDL performs better in detecting the sensor-based threats compared to the state-of-art existing methods by attaining the accuracy of 99% in identifying benign and malicious activities, (d) SBTDDL also detects the malicious activity occurring like benign activity, and the performance is not affected when the total number of benign and malicious activities increases.