Enhancing Security in Healthcare Frameworks using Optimal Deep Learning-based Attack Detection and Classification for Medical Wireless Sensor Networks

Wireless Sensor Networks (WSNs) have modernized healthcare, providing vital sign collection and real-time patient monitoring. Healthcare WSNs are vulnerable to cyberattacks, such as false data injection, sensor manipulation, and data eavesdropping, which can disrupt monitoring and endanger patient lives. Traditional Intrusion Detection Systems (IDSs) based on static signatures struggle with evolving threats. Deep Learning (DL)-based IDSs, combined with Feature Selection (FS), offer a more adaptive and effective solution, improving attack detection and protecting patient data. This work presents an innovative Pigeon-Inspired Optimizer-based Feature Selection with Deep Learning-based Attack Detection and Classification (PIOFS-DLADC) method, which focuses on creating an optimal DL framework for attack detection and classification in healthcare WSNs. Initially, patient health data (actual input data) undergo preprocessing using the one-hot encoding system. Then, the PIOFS method selects key features from sensor data streams, reducing dimensionality and improving model efficiency. Furthermore, an attention-based Bidirectional Gated Recurrent Unit (BiGRU) method captures long-term dependencies and prioritizes features for accurate attack classification. The Coati Optimization Algorithm (COA) is employed to tune the hyperparameters of the DL models. The model efficiently explores the hyperparameter space, optimizing the performance for attack detection and classification. Validated on a healthcare WSN dataset, the PIOFS-DLADC model demonstrated an accuracy of 96.78%, which is superior to existing approaches.