Improving Model Explainability in AD Prediction: SHAP-Based Feature Attribution and Interpretable Ensembles

Considering the growing number of Alzheimer’s disease (AD) patients, there is an increasing demand for early and reliable detection methods. While numerous Machine Learning (ML) and Deep Learning (DL) models have been developed for AD diagnosis, their clinical adoption is hindered by limited interpretability. Most current studies emphasize model performance but overlook explainability, which is crucial in medical decision-making. This study explores the interpretability of a Decision Tree (DT), a Sequential Neural Network (SNN), and their ensemble (DT-SNN) using SHapley Additive exPlanations (SHAP), an explainable artificial intelligence (XAI) technique. By analyzing and comparing SHAP-based feature attributions across models, we aim to enhance transparency, uncover relevant risk factors, and assess the fidelity of model predictions. This approach supports the development of explainable, trustworthy models that bridge the gap between predictive accuracy and clinical insight in AD detection.