ScottishFold: CatBoost-Enabled Lightweight Autonomous Smart Home Device Classification
Tawan Hohum, Kalika Suksomboon, Aimaschana Niruntasukrat, Sophon Mongkolluksamee, Natapon Tansangworn · 2021 IEEE Globecom Workshops (GC Wkshps) · 2021
Massive deployed IoT devices in smart homes pose security management challenges due to different hardware and firmware. To address these challenges, an automated IoT-device-type identification has been used as the first step in autonomous networks in which IoT security policy is automatically managed. Although machine-learning-based traffic patterns were proven high accuracy on a prediction while heavy on preprocessing, we argue that good stability and less computational cost should be considered. We focus on constructing an IoT-device-type classification based on the common features extracted from the network traffic. Inspired by the categorical boosting algorithm, we propose ScottishFold, a classification model, that is constructed from the dirty categorical IoT traffic features with less preprocessing. ScottishFold can classify the IoT device type from a single packet (either a signaling handshake or a data packet) that makes the classification process light. We boost the classification accuracy by integrating a machine reasoning approach into the classification process. The experimental results confirm the effectiveness and robustness of our proposed technique.