Abstract
Smart home networks are increasingly vulnerable to cyberattacks due to the rapid expansion of Internet of Things (IoT) devices with limited computational capacity. Traditional intrusion detection systems are often too resource-intensive for edge environments, highlighting the need for lightweight, local solutions. This study designs and evaluates an efficient anomaly detection system for smart home network traffic using machine learning techniques. We use the IoT-23 dataset for evaluation and presents a detection framework that integrates three lightweight models — Random Forest (RF), Isolation Forest (IF), and a deep Autoencoder (AE) — implemented in Python via scikitlearn and TensorFlow to provide multi-layered defence at the edge. A prototype system is developed on a Raspberry Pi (via FastAPI), tailored for edge deployment on low-power gateways. Experimental results show that the RF achieves the highest overall detection accuracy, while the AE and IF are effective at detecting novel or unlabeled anomalies. The proposed approach provides a scalable, privacy-preserving detection method suitable for smart home environments. This research contributes to the development of lightweight, real-time cybersecurity solutions for edge IoT systems and lays the groundwork for future integration with adaptive and federated learning frameworks.