When machine learning meets medical data : pitfalls, insights, enhancements

Abstract

This thesis aims to investigate the intersection of machine learning and medical data by identifying potential challenges, gaining crucial insights, and exploring methods for improvement. Despite the increasing popularity of machine learning in medical data analysis, several challenges still need to be addressed. Therefore, a Monte Carlo simulation study is conducted to examine the challenges related to feature selection, sample size, and classifier selection, which can result in issues such as poor labeling performance.

To address these challenges, a series of case studies and experiments are conducted. The results illustrate that incorporating label dependency can effectively enhance machine learning models, particularly in cases of extreme multi-label learning (XML) problems characterized by highly skewed data distributions. These enhancements significantly improve model performance and emphasize their importance in tackling the challenges associated with machine learning in medical data analysis.

Overall, this thesis contributes to the field of medical data analysis by providing insights into the challenges and opportunities associated with machine learning. By exploring methods for improvement and addressing potential pitfalls, this thesis aims to facilitate the development of accurate and reliable machine learning models for medical data analysis, ultimately leading to improved healthcare outcomes.