An integrated machine learning and deep learning framework for intrusion detection in IoT smart homes

The widespread adoption of Internet of Things (IoT) devices in smart homes has introduced new vulnerabilities, making these environments increasingly susceptible to cyber threats. Firewalls and traditional intrusion detection systems (IDS) are often insufficient in addressing sophisticated attacks. The proposed hybrid machine learning approach overcomes the limitations of traditional systems for intrusion detection in IoT smart homes. The framework integrates supervised learning, deep learning, and ensemble techniques to enhance accuracy and reduce false positives. The proposed approach is evaluated using UNSW-NB15, BOT-IOT, and real-world Wireshark IoT intrusion detection datasets. Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), Convolutional Neural Network (CNN), and a hybrid votning classifier (DT + RF + AdaBoost) are employed, where the hybrid model serves as the ensemble component of the broader framework. Experimental results show that CNN achieved approximately 99% detection accuracy, while the hybrid model attained up to 99.8% accuracy on the BOT-IoT dataset. Across multiple datasets, the hybrid model consistently achieved high performance, with 99.57% accuracy on UNSW-NB15 and 99.75% on the Wireshark IoT dataset, outperforming traditional classifiers. Additionally, dimensionality reduction techniques such as Maximal Information Coefficient (MINE) and Principal Component Analysis (PCA) were employed to reduce computational complexity without sacrificing accuracy. This study underscores the effectiveness of hybrid AI models for intrusion detection in IoT smart homes and recommends future directions including federated learning and zero-day threat detection.