A NOVEL INTRUSION DETECTION MODEL FOR CRITICAL HEALTHCARE ENVIRONMENTS

In this study, a deep learning-based model was proposed by combining a sparse autoencoder and a combination of autoencoders with LSTM for feature selection and intrusion detection. Subsequently, the likelihood of attacks was evaluated using an ensemble method. The proposed model employed several functions and addressed current research gaps, such as reducing false positive rates, mitigating model overfitting, addressing data imbalance, and identifying new attack scenarios. The proposed model was tested on benchmark datasets, viz., WUSTL-EHMS 2020, IoT Healthcare Security 2021, and CIC IoMT 2024. The proposed model achieved a perfect detection rate in binary classification of WUSTL-EHMS 2020 and IoT Healthcare Security 2021. The model achieved an accuracy rate of 99.80% (binary) and 90.67% (multiclass) in the CIC IoMT 2024 dataset. In addition, the Matthews Correlation Coefficient (MCC), Cohen’s Kappa, and Adversarial Robustness Score (ARS) provided a comprehensive assessment of the model, demonstrating its robustness and applicability in healthcare.