IoMT Intrusion Detection System: A CNN-LSTM Hybrid with Adaptive Learning

IDS is being used essentially for network security which also extends its usage in Internet of Medical Things. Conventional IDS methods are marred by the high computational cost, lack of scalability, and vulnerability to the class imbalance issue. A new CNN-LSTM based hybrid model with integrated knowledge distillation and pruning methods is developed for effective yet timely IoMT intrusion detection. The IDSNet for IoMT approach utilizes a teacher-student approach, where a higher-level CNN-LSTM teacher compresses its own knowledge into a slender student model in an effort not to do much redundant computations and to have effective detection. SMOTE addresses class imbalance, and polynomial decay pruning reduces resource utilization without losing much accuracy. Compared to conventional methods on the CIC-Malmem dataset, the approach outperforms the existing methods and strikes a balance among accuracy, efficiency, and deployability. The model performance for IoMT intrusion detection is emphasized to provide an encouraging solution to guaranteeing network integrity in practical IoMT networks.