PREDICTION OF ALZHEIMER’S DISEASE USING HYPERPARAMETER-TUNED XGBOOST

Abstract

This study introduces an effective machine learning framework for the early and accurate prediction of Alzheimer’s Disease (AD) using tabular clinical data. The proposed model is built on the Extreme Gradient Boosting (XGBoost) classifier, fine-tuned for high performance and strong generalization. To address class imbalance commonly found in clinical datasets, the Synthetic Minority Over-sampling Technique (SMOTE) was applied only to the training data. Model optimization was carried out using Grid Search Cross-Validation (Grid Search CV) to identify the best hyperparameter configuration. When evaluated on an independent test set of 800 samples, the model achieved impressive results—98.75% accuracy, 98.28% precision, an F1-score of 0.9856, and an AUC of 0.9803. Feature importance analysis revealed that Smoking Status, Functional Assessment Scores, and Diabetes were the most influential predictors. Overall, the findings highlight the model’s strong predictive capability and its potential as a reliable, non-invasive diagnostic support tool for Alzheimer’s Disease.