Intrusion Detection System-based DNN+GRU-MHA for Securing Medical and Industrial IoT Environments

The growing prevalence of medical and industrial IoT in various domains has intensified the need for robust security mechanisms, particularly real-time systems with minimal false positive rates (FPR). Several solutions have been proposed to address these challenges, with artificial intelligence (AI)-based intrusion detection systems (IDS) emerging as a promising approach. This study focuses on developing a novel hybrid model that integrates Deep Neural Networks (DNN), Gated RecurrentUnits (GRU), and a Multi-Head Attention (MHA) mechanism. The proposed DNN+(GRU-MHA) model is trained on two distinct datasets: CICIoMT2024 for the medical domain and EdgeIIoTset for the industrial domain. To mitigate class imbalance, RandomOverSampler is applied to the medical dataset, while Synthetic Minority Over-sampling Technique (SMOTE) is used for the industrial dataset. Experimental results demonstrate the model’s superior performance, achieving accuracy rates of 98.22% and 99.78% for medical and industrial datasets, respectively. The model also attained exceptionally low FPR values of 0.002% and 0.0000% in medical and industrial datasets, respectively, with inference times of 0.0001 seconds per instance. These results validate the effectiveness of the proposed model, demonstrating its capability to operate in diverse environments and detect various types of cyber threats.