Machine Learning-Based Framework for Detecting Unauthorized IoT Devices

The widespread adoption of Internet of Things (IoT) devices across homes and enterprises has introduced significant security risks, especially when unauthorized or compromised devices gain access to sensitive networks. This paper proposes a machine learning-based framework to detect unauthorized IoT devices in real-time using features extracted from TCP/IP traffic. We utilize a Random Forest classifier trained on labeled network traffic from authorized devices. The proposed approach detects device types not on a pre-established whitelist, achieving an average of 96% accuracy in identifying unauthorized devices based on a 20-session window classification. The framework generalizes across different vendors, supports real-time alerting, and is resilient against adversarial attacks.