Intelligent Intrusion Detection System for IoT Networks: A Feature Enhanced Deep Belief Network Approach

This study presents a novel and resilient Intrusion Detection System (IDS) designed to address the various challenges posed by Internet of Things (IoT) systems. The proposed model utilizes a Deep Belief Network (DBN) for feature selection in conjunction with Mean Shift clustering, enhancing detection accuracy. Additionally, the system utilizes adaptive drift detection to efficiently manage high-dimensional and dynamic IoT networks. Evaluation of the IoT20 dataset demonstrates that the proposed DBN IDS achieves a Root Mean Square Error (RMSE) of 0.025, with a precision of 97.8% and a recall of 98.1%. The model also boasts an F1 score of 97.9% and a specificity of 98.7%. Qualitative analysis reveals an Area Under the Curve (AUC) of 0.98 for the Receiver Operating Characteristic (ROC) curves, indicating near-perfect classification performance. Other performance metrics of the IDS are favourable, with reconstruction error limiting factors reported at 0.022, suggesting a reliable model for normal behaviour and accurate anomaly detection. The model’s accuracy for inference and evaluation stands at 96% and 97%, respectively, illustrating the consistency of the studied model across different data segregations. These results highlight that the proposed DBN-based IDS is a reliable solution that goes beyond current threats in the IoT landscape, effectively detecting both existing and emerging threats.