Early Detection of Alzheimer's Disease Using Explainable Machine Learning on Clinical Biomarkers: A Multi-Class Classification Study Using the Alzheimer's Disease Neuroimaging Initiative (ADNI) Dataset

Researchers developed an explainable machine learning model using XGBoost to detect Alzheimer's disease stages from routine clinical assessments, achieving 98.2% accuracy on three-class classification (normal cognition, mild cognitive impairment, and Alzheimer's disease). The model uses SHAP analysis to provide interpretable feature importance, identifying clinical biomarkers like CDR Global and MMSE as key predictors.

This study demonstrates the application of interpretable machine learning to a significant healthcare challenge affecting 55 million people worldwide. The researchers constructed an XGBoost classifier on 1,641 baseline subjects from the ADNI dataset, optimizing hyperparameters and addressing class imbalance through SMOTE. The model achieved near-perfect performance metrics: 98.2% macro AUC, 94.3% accuracy, and 0.909 Cohen's kappa on the held-out test set.

The research represents a broader trend in medical AI where model interpretability increasingly matters alongside predictive accuracy. Traditional black-box deep learning models face adoption barriers in clinical settings due to regulatory and trust concerns. By employing SHAP values, the authors revealed clinically plausible feature importance patterns—CDR Global dominates normal cognition and MCI detection, while CDR-SB and MMSE together drive Alzheimer's classification. This alignment with clinical knowledge enhances confidence in the model's validity.

For healthcare stakeholders, this approach offers immediate practical value. Routine clinical assessments (MMSE, CDR, MoCA, FAQ) already exist in standard practice, eliminating implementation barriers. The explainability component addresses physician skepticism, making AI integration into diagnostic workflows more feasible. Insurance companies and healthcare systems could reduce diagnostic delays and standardize assessment protocols.

Future development directions include multimodal detection incorporating speech biomarkers, potentially capturing linguistic and cognitive decline patterns undetectable through traditional assessments. Early detection frameworks like this could shift dementia care from symptomatic management toward preventive intervention strategies, though validation on diverse populations remains essential before widespread deployment.

• →XGBoost model achieves 98.2% accuracy detecting three Alzheimer's disease stages using only routine clinical features
• →SHAP explainability analysis reveals CDR Global as primary predictor for normal cognition and mild cognitive impairment classification
• →Model trained on 1,641 subjects shows clinical plausibility with class-specific feature importance patterns
• →Routine clinical assessments eliminate implementation barriers compared to expensive neuroimaging biomarkers
• →Future work plans integration of speech biomarkers for enhanced multimodal dementia detection