Explainable artificial intelligence for early Alzheimer’s diagnosis using enhanced grey relational features and multimodal data

Abstract Alzheimer’s disease, a progressive neurodegenerative disorder, presents a growing global health challenge due to its increasing prevalence and lack of accessible early diagnostic methods. Even though it has enhanced the diagnostic accuracy of machine learning, there is a major concern about striking a balance between predictive performance and interpretability. The proposed study presents an interpretable and sustainable machine learning architecture for early diagnosis of Alzheimer’s disease based on multimodal, structured clinical and behavioral data, including demographics, vascular risk factors, lifestyle, and cognitive data. We perform extensive feature engineering to derive composite features, including blood pressure ratio, MMSE age ratio, cholesterol ratio, and cognitive decline score. The class imbalance is addressed using the Synthetic Minority Oversampling Technique. We also introduce a new strengthened Grey Relational Grade index based on the theory of grey system and the policy of sigmoid normalization. This greatly enhances the feature-diagnosis correlation (0.725 to 0.891), representing complicated nonlinear associations. This paper compared seven mainstream classifiers, such as Logistic Regression, Random Forest, Extreme Gradient Boosting, Light Gradient Boosting Machine, CatBoost, Stacking Ensembles, and Deep Neural Networks, in the context of model comparison. Among them, Deep Neural Networks achieve the best performance (accuracy: 98.01%, AUC: 99.43%), followed by a CatBoost-based Stacking Ensemble (Accuracy: 97.91%, AUC: 98.10%). Shapley Additive Explanations make models easier to understand by showing important modifiable predictors like family history, smoking, and early cognitive symptoms. This study presents that combining enhanced Grey Relational Grade metrics with robust machine learning and deep learning models produces an accurate, interpretable, and potentially effective framework for early AD risk assessment, which can be used to implement effective, behavior-centric prevention strategies in ageing demographics.