An Explainable AI Approach to Alzheimer’s Prediction using Ensemble Methods on Imbalanced Data

Alzheimer’s disease (AD) is a progressive neurodegenerative condition that necessitates early detection to enable effective interventions and slow its progression. This research presents an explainable machine learning approach using ensemble classifiers to predict Alzheimer’s risk based on demographic, clinical, cognitive, and lifestyle factors. Mutual Information (MI) was employed for feature selection, and models using Extra Trees, XGBoost, and AdaBoost classifiers were developed. The findings of this study reveal that ensemble methods, particularly XGBoost, significantly outperform individual models in predicting Alzheimer’s disease risk. The dataset, sourced from Kaggle, includes 2,149 patient records with 33 input features. After comprehensive preprocessing, which involved managing outliers, scaling, and oversampling for minority classes, the data was split into training and testing sets in an 80:20 ratio. Performance was evaluated using metrics such as accuracy, precision, recall, F1 score, and ROC-AUC. Among the models, XGBoost achieved the highest performance, with a test accuracy of 92.1%, a recall of 90.0%, and an ROC-AUC score of 96.7%, with AdaBoost and Extra Trees closely following. The confusion matrix indicates that XGBoost had the most true positives (271) and the fewest false negatives (30), highlighting its superior reliability in identifying positive AD cases. To enhance model transparency, SHAP (SHapley Additive Explanations) was used to analyze feature importance. SHAP summary and bar plots identified the top predictors. Overall, this study demonstrates the effectiveness of ensemble classifiers, especially XGBoost, in predicting Alzheimer’s and emphasizes the importance of Explainable AI for clinical trust. The use of SHAP improves interpretability, making the proposed system a practical and transparent tool for healthcare professionals aiming for early detection of Alzheimer’s risk.