Automatic alzheimer's disease diagnosis using XGBoost with SHAP for explainability

Abstract

Alzheimer's disease (AD) is a progressive neurodegenerative disorder that is challenging to diagnose early, with traditional methods often being time-consuming and costly. Early detection is essential for effective treatment, but more accessible and automated methods are needed. Current approaches rely on clinical assessments and neuroimaging, which lack the efficiency and cost-effectiveness required in clinical practice. While machine learning has shown promise in classification, achieving both high accuracy and interpretability remains a challenge. To resolve this, we present an automated method using the XGBoost classifier with SHAP (Shapley Additive exPlanations) for model interpretability, incorporating Explainable AI (XAI). Our model preprocesses clinical and demographic data, enhances feature extraction with shape-based techniques, and applies XGBoost for classification. SHAP is used to explain predictions and identify influential features, ensuring transparency in the decisionmaking process. We utilize Explainable AI (XAI) to ensure that the model's decisions are transparent, allowing healthcare professionals to understand the reasoning behind predictions. It has been observed that our model achieves high accuracy and provides clear insights into the key features driving predictions. By incorporating shape-based features and visualizations such as SHAP summary, dependence, and force plots, the model's interpretability is further enhanced. This approach offers a promising, cost-effective tool for early AD detection, with advantages in both performance and transparency over traditional methods.