An Explainable Machine Learning–Based Web Tool for Alzheimer’s Disease Diagnosis

Abstract

Alzheimer’s disease (AD) is one of the most critical global health challenges, and early diagnosis remains essential for slowing disease progression and improving patient management. This study proposes an automated and interpretable machine learning framework for AD diagnosis that combines feature selection, explainable artificial intelligence (XAI), and deep learning approaches to support clinically transparent decision-making. A publicly available AD dataset containing 25 attributes was used. Feature importance was assessed using correlation analysis together with explainability-based importance measures derived from SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME), leading to the identification of an optimal subset of features composed of Age, presence of the GeneticRiskFactorAPOE-ε4 allele and FamilyHistoryOfAlzheimer. Several classical machine learning models, including Logistic Regression (LR), K-Nearest Neighbors (KNN), Decision Tree (DT), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost) were evaluated alongside Deep Learning (DL) approaches based on Multilayer Perceptron (MLP) architectures. Two MLP-based models were investigated: a feature selection-based MLP (FS-MLP), trained on the selected features, and an encoder-based MLP, integrating deep learning-driven dimensionality reduction.SHAP and LIME were further employed to provide both global and local explanations of model predictions. {The selected model} was deployed in a web-based application enabling AD diagnosis and result interpretation.Performance evaluation using cross-validation, confidence intervals, and statistical significance testing showed that tree-based models achieved the best overall performance, with FS-XGBoost obtaining the highest Area Under the Curve (AUC) of 80.1%, a recall of 77.8% and a specificity of 68.8%, indicating a balanced sensitivity–specificity trade-off for screening applications.To enhance practical applicability, the selected model was integrated into a user-friendly web-based decision-support application for AD diagnosis and interpretation. The results demonstrate that combining feature selection, explainable AI, and machine learning techniques, can provide an effective and transparent framework for supporting early AD diagnosis in clinical settings.