Deep Learning for Anomaly Detection in IoT Healthcare Systems

This study introduces a new hybrid deep learning method for intrusion detection in the Internet of Medical Things (IoMT), a rapidly expanding domain that enhances patient care but remains highly vulnerable to cyber threats. The increasing integration of IoMT devices in healthcare facilitates real-time monitoring and data exchange, yet their susceptibility to security breaches poses serious risks to patient privacy and system integrity. As these devices generate vast amounts of sensitive data, ensuring security against cyberattacks is critical. Our proposed method integrates an Autoencoder (AE) with three encoder-decoder layers for anomaly detection and a Long Short-Term Memory (LSTM) network for temporal analysis. The autoencoder identifies anomalies through reconstruction errors and latent space classification, while the LSTM network captures sequential patterns in network traffic to detect attack signatures. We evaluated the model using the CICIoMT2024 data set, which includes traffic from 40 IoMT devices and 18 distinct attack types across Wi-Fi, MQTT, and Bluetooth protocols. The data set presents a significant class imbalance, with DoS and DDoS attacks dominating, posing real-world security challenges. To address this, we employed data balancing techniques to improve model performance. Our evaluation shows that the hybrid model achieves 94.1% accuracy with a robust Area Under the Curve (AUC), significantly outperforming the Autoencoder alone. Our findings demonstrate the efficacy of employing deep learning techniques to bolster IoMT security. This approach enables swift identification of various cybersecurity threats and establishes a resilient defense system against emerging attacks.