Advancing IoT Security: A Stacked Hybrid AI Approach for Anomaly Detection
D. Ananya, Kasireddy Mahalakshmi, Pallavi Joshi · 2024
AI-driven anomaly detection for Internet of Things networks is a major challenge in security. A range of ML models and DL are used in this field. Fascinatingly, multi-layer perceptron (MLP) and Long short-term memory (LSTM) models attain 92.38% accuracy, while Naive Bayes and DT models reach 93.07%. These findings provide important information about the relative benefits of different strategies and demonstrate how hybrid approaches may be used to bolster IoT security against emerging threats. Motivated by this, we proposed a hybrid AI model for anomaly detection in various scenarios of IoT networks. The work introduces stacked hybrid models that incorporate RF, DT, and Naive Bayes. The model is trained and tested on the UNSW-NB15 dataset and an impressive accuracy of 94.94% is achieved. Additionally, there is a reasonable 78.97% accuracy for both the Restricted-Boltzmann Machine (RBM) and Recurrent Neural Network (RNN) models. The results show that our proposed hybrid AI anomaly detection model has performed exceptionally well by achieving high accuracy, precision, recall, and F1-score compared to the existing implementations.