XGBoost spots early Alzheimer's from routine clinical tests
Researchers trained an XGBoost classifier on eight clinical biomarkers from the ADNI dataset to distinguish normal cognition, mild cognitive impairment, and Alzheimer's disease.
• Input features: MMSE, CDR Global, CDR-SB, MoCA, FAQ, age, sex, education
• SMOTE balancing and Optuna hyperparameter search (50 trials)
• Evaluation: macro AUC-ROC with 1,000-iteration bootstrap CI, macro F1, balanced accuracy, Cohen's kappa
• SHAP values provide feature-level interpretability for clinical decision support