ANALYSIS AND ASSESSMENT OF IMBALANCE DATA FOR PREDICTIVE MODELING IN HEALTHCARE

Abstract

The Internet of Medical Things (IoMT) is transforming modern healthcare by enabling real-time monitoring, remote diagnosis, and smarter clinical decision-making. In our first study, we explored the evolving landscape of IoMT, highlighting its potential to improve patient outcomes through intelligent devices that collect and transmit medical data. While the adoption of IoMT is growing rapidly, one of the major challenges we identified is the presence of imbalanced and irrelevant data. These data issues can sig- nificantly impact the accuracy of critical healthcare decisions, especially when machine learning models are used to detect anomalies or predict patient conditions. To address this challenge, our second study presents an enhanced machine learning framework specifically designed to improve software defect prediction by handling imbal- anced datasets more effectively. We introduced a refined version of the ASRA model, replacing the traditional Chi-square method with a hybrid feature selection approach us- ing ReliefF and Information Gain. Additionally, we applied a combination of SMOTE and Tomek Link techniques to balance the dataset while reducing noise. A cost-sensitive AdaBoost classifier, using the J48 decision tree as the base learner, further improved the model’s ability to identify rare but critical instances. By connecting these two works, this thesis aims to bridge the gap between the tech- nical advancements in software reliability and the practical challenges in healthcare IoT applications. Our approach not only enhances the reliability of data-driven systems in IoMT but also contributes to safer and more effective healthcare technologies.