An efficient and explainable machine learning framework for Alzheimer's disease detection

Alzheimer's disease detection is of paramount importance as it is a global health problem in terms of early detection, interventions, and management. The study applies prominent machine learning classification to Alzheimer's disease by integrating longitudinal and cross-sectional data analysis. With careful correlation analysis, we derived the following important predictive features: 'M/F', 'Age', 'Education', 'SES', 'CDR', 'MMSE', 'eTIV', 'nWBV', 'ASF', and 'Hand'. In this study, the entire dataset was divide into training (80%) and testing (20%) sets, with eight machine learning models implemented on it for classification: CatBoost-CB, Logistic Regression-LR, Extreme Gradient Boosting-XGB, Support Vector Machine-SVM, Random Forest-RF, K-Nearest Neighbors-KNN, Decision Tree-DT, and Gradient Boosting-GB. Our results demonstrated superior performance from ensemble models, with RF, XGB, GB, DT, and CB achieving 98.1% accuracy, 98.2% precision, and 98.1% recall. These models showed exceptional performance with a Kappa score of 0.95, indicating strong reliability in classification. We have also used LIME explainability to enhance model transparency, providing insight into the contribution of different features in support of the interpretability of the model. In this study, we stress the power of machine learning for early detection of Alzheimer's disease, and provide future research directions on improving the predictive accuracy and transparency of predictive models.