Random Forest Based Dual Attack Detection for Smart Internet of Things (IoT) Home Appliance
With the rapid proliferation of IoT technology, smart home appliances face growing data security threats, especially model probing attacks and network-layer DDoS attacks. This paper proposes a machine learning based dual detection system. For model probing defense, we design a query-behavior detection framework extracting three core indicators (top‑1 probability, margin, top‑2 probability) and employ a Random Forest classifier. For IoT network attacks, we construct an end‑to‑end detection framework using multi‑dimensional traffic features. Experiments on two public datasets show that the probing attack detector achieves 75.7% and 82.1% accuracy on DDoS_1 and DDoS_2, while the network attack detector achieves 76.53% accuracy and 0.76 F1‑score. Feature importance analysis reveals key detection signals. We discuss performance bottlenecks and future optimization directions including data balancing, deeper feature engineering, deep learning models, and lightweight edge deployment.