Enhanced Deep Learning Approaches for Intrusion Detection in Iot Healthcare Systems

Abstract

The rapid advancement of Internet of Things (IOT) technology has newlinetransformed healthcare by improving connectivity and enabling real-time patient data newlinemonitoring. However, the proliferation of IoT devices has introduced significant newlinesecurity vulnerabilities, necessitating robust intrusion detection to protect sensitive newlinehealthcare information from unauthorized access and malicious threats. Current newlineIntrusion Detection Systems (IDS) face challenges in addressing the dynamic, resourceconstrained newlinenature of IOT environments, resulting in inefficiencies in detecting and newlineclassifying malicious activities. This research aims to develop an optimized IDS newlinespecifically designed for IOT-based healthcare applications, capable of accurately newlineidentifying and categorizing complex threats while addressing the limitations of newlineexisting solutions. The thesis is organized into two major phases, each presenting newlineadvanced IDS methodologies that leverage deep learning and optimization techniques newlineto enhance security in IoT-driven healthcare environments. newlineIn Phase 1, an improved IDS model is developed using Adaptive Generative newlineAdversarial Networks (AGAN) for dynamic feature selection, enabling the system to newlinefocus on the most relevant data for effective intrusion detection. Additionally, a Support newlineVector Machine-based FOX optimization algorithm (SVM-FOX) is employed to newlineclassify network activities as either normal or malicious, with further sub classification newlineof malicious activities into nine specific categories, including Shell code, Exploit, newlineBackdoor, Reconnaissance, and Denial of Service (DoS). By optimizing SVM newlineparameters with the FOX algorithm, this approach achieves high accuracy in newlinedistinguishing normal from malicious network traffic. Experiments conducted on the newlineUNSW-NB15 dataset demonstrate the effectiveness of the SVM-FOX model, with newlineassessments based on Kappa score, Matthew Correlation Coefficient, precision, F1 newlinescore, recall, and accuracy. Results indicate that SVM-FOX outperforms existing newlinemethods in both detection accuracy and computat