SHAP-based Feature Selection and Explainable Machine Learning Classification of Alzheimer's Disease

Abstract

Alzheimer's Disease (AD) is a progressive neurodegenerative disease with a significant impact on healthcare. This study suggests a machine learning (ML)-based Explainable Artificial Intelligence (XAI) framework for early AD prediction. The system incorporates SHAP (Shapley Additive explanations) for interpretability in order to overcome the "black box" aspect of existing machine learning models. Using the OASIS dataset, we quickly investigate different classifiers for AD prediction by utilizing PyCaret, a low-code tool. With a classification accuracy of 96%, the Naive Bayes classifier was the most successful of the assessed models. In order to comprehend feature importance and get insights into model thinking, SHAP analysis is applied. The approach also selects features, determining the most important variables for AD prediction by utilizing SHAP values. The model's transparency is increased by this combination method, which uses SHAP for interpretability and feature selection and PyCaret for effective exploration. Clinicians gain deeper understanding of the model's decision-making process and the factors most critical for AD prediction. This study breaks new ground by demonstrating the efficacy of PyCaret and SHAP in building an interpretable and accurate framework for early AD prediction.