Advancing Feature Selection and Interpretability in Machine Learning Models for Medical Data Analysis
Author: Irem Ozturk
Faculty Supervisor: Shahrukh Humayoun
Department: Computer Science
This research project focuses on enhancing feature selection and interpretability in machine learning models for medical data analysis. Six different AI gradient boosting methods, decision trees, random forests, support vector machines, generalized additive models (GAM), and linear models are employed to identify critical attributes impacting model predictions across diverse medical datasets. Utilizing convolutional neural networks (CNN) for evaluation, partial dependence plots (PDPs) are utilized to visualize feature impacts, facilitating insights crucial for clinical decision-making and patient care.