Drift-Aware Resilient Ensemble Deep Learning-based Anomaly Detection Model in Internet of Medical Things

This is the authors' abstract. We don't add key points for this paper.

Background: The rapid advancement of the Internet of Medical Things (IoMT) has facilitated continuous health monitoring and real-time physiological data collection. Artificial intelligence, particularly machine learning and deep learning techniques, has become widely employed for anomaly detection. Methods: This research proposes a novel Drift-Aware Resilient Ensemble Deep Learning-based Anomaly Detection Model in Internet of Medical Things Environments (DREAM-IoMT). The proposed method employs a hybrid feature selection strategy integrating Gini-index, recursive feature elimination, and SelectFromModel. For anomaly detection and classification, the proposed DREAM-IoMT model employs a weighted-voting ensemble of a graph convolutional network, a bidirectional temporal convolutional network, and an improved conditional variational autoencoder. This ensemble approach enhances accuracy by accounting for each classifier’s prediction confidence. The ensemble classification model is further optimized using AdamW. Results: The DREAM-IoMT model achieved an accuracy of 98.74%, demonstrating superior anomaly detection performance compared with existing methodologies. Conclusion: The proposed DREAM-IoMT effectively combines feature selection and weighted ensemble deep learning for reliable anomaly detection in IoMT environments.