COMPUTER SCIENCE
Permanent URI for this collection
Browse
Recent Submissions
Item ADAPTATION OF XAI-FML METHODOLOGY FOR EFFICIENT AND SECURE E-HEALTHCARE SYSTEM(KINNAIRE COLLEGE COMPUTER SCIENCE DEPARTMENT, 0022-06-23) RABIA ABIDArtificial Intelligence (AI) has been applicable in many sector (like educations, healthcare, businesses, government bodies, etc) to lessen the human effort and to create an ease. All AI based systems have decision support systems (DSS) to help human in all high-pitch and low pitch situation. As a support system many machines learning (ML) based algorithms helps to make accurate decision according to situation, increase accuracy rate in data classification and enhance the performance of systems. Explainable AI (XAI) has advance feature to en hanced the decision-making feature and improve the rule base technique by using more ad vance ML and deep learning (DL) based algorithms. In this research we chose e-healthcare systems for efficient decision making and data classification, where quite massive data like patients’ health record (PHR), hospitals confidential information, administrative data, research data, physicians’ details, and many other. In this research work, we identify the existing gaps in traditional AI and ML based algorithms for efficient e-healthcare systems and trying to over come it by using XAI and advance ML based algorithm Federated Machine Learning (FML). FML is a new and advance technology which helps to maintain privacy for PHR and handle large amount of medical data effectively. In this context, XAI along with FML increase the efficiency and improve the security also of the e-healthcare systems. The performed experi ment shows the efficient system performance by implementing federated averaging algorithm on open source FL platform. The evaluating graphs shows the accuracy rate by taking epochs size 5, batch size 16 and no. of clients 5, which shows higher accuracy rate with (19, 10−4). To conclude our research, we discuss the future work with still existing some gaps in e-healthcare system like security, price, efficiency, performance evaluation and many other.Item Intrusion Detection in IoMT based Smart Healthcare System (SHS) using Deep Learning techniques(KINNAIRE COLLEGE COMPUTER SCIENCE DEPARTMENT, 0022-06-23) Attiya KhanThe Internet-of-Things (IoT) has infiltrated nearly every aspect of life. One of the most important areas where IoT solutions and infrastructures are used widely is smart healthcare system (SHS). IoT-based smart healthcare solutions have significantly increased the benefit of the healthcare sector with the use of mobile and wearable devices. Smart healthcare reduces hospitalization costs and provides timely treatment for a number of medical conditions by incorporating IoT sensors into health monitoring equipments. Today, the purpose of healthcare systems is not confined to treating patients only. In SHS, wearables, implantable devices, and sensors monitor the vital parameters of a patient. These parameters are sent for evaluation to the emergency services or healthcare professionals. This results in a significant usage of health data exchange for improved, timely, and more accurate diagnosis. Nevertheless, SHS are extremely prone to a variety of security breaches and malicious attacks, such as tampering, privacy leakage, and forgery. In the smart healthcare domain, it is essential to take a systematic approach to privacy and security measures in communication, data storage, interconnecting things, and data handling. In various studies, several intrusion detection systems (IDS) have been proposed to detect cyber security threats in SHS and to identify malicious attacks and privacy breaches. This study was conducted as a consequence of the limits of IDSs in responding to challenges and attacks and in implementing attacks and privacy access control in the SHSs. In this study, we designed a deep learning-based intrusion detection system to efficiently identify smart healthcare network intrusions by evaluating traffic flow data. We specifically used the “Long Short-Term Memory (LSTM)” technique to detect malicious attacks and other security threats in SHS. In this system, we utilized the CFS algorithm for feature selection. The objective is to select a subset of features having a high feature-class correlation in order to maintain or boost predictive power, and low feature-feature correlation to prevent redundancy. We evaluated the proposed system using Wustl-ehms-2020 IoMT dataset. The proposed system achieves accuracy of 96%, which is greater than existing v approaches. This study demonstrates that our approach outperforms other cutting-edge techniques for intrusion