Active Learning for Detecting Hardware Sensors-Based Side-Channel Attack on Smartphone

The evolution of smartphones made people more dependent on them since they save their data on smartphones for convenience. This convenience aroused threats to the stored data through cyber attacks. Smartphones consist of various hardware sensors, including a magnetometer, an accelerometer, and a gyroscope, which can be vulnerable to cyberattacks such as side-channel attacks. These attacks can deduce the typed keystrokes of a soft keyboard through the vibration they produce; it can cause data leakage and threaten user privacy. This paper proposes a novel approach to detecting side-channel attacks that combine active learning-based machine learning models: Random Forest, eXtreme Gradient Boosting, Decision Tree, Gradient Boosting, K-Nearest Neighbors, and Light Gradient-Boosting Machine classifier. The approach is evaluated on a dataset generated through the smartphone’s hardware sensors (i.e., accelerometer, gyroscope, and magnetometer). The performance is evaluated in three rounds; evaluation measures demonstrate that the keystrokes can be inferred with an accuracy of 91.99% and F1-score of 91.89%. The outcomes indicate that this approach is highly accurate and effectively increases the inference rate to detect attacks compared to conventional ML techniques and existing research.