AlzStack: Forecasting early-onset Alzheimer's with an explainable AI system using multiple data balancing techniques.

Area of Science:

• Artificial Intelligence in Medicine
• Neurology
• Machine Learning for Healthcare

Background:

• Alzheimer's disease (AD) diagnosis relies on methods that are often time-consuming, expensive, and inconsistent.
• Early detection of AD is critical for timely intervention and improved patient prognosis.

Purpose of the Study:

• To develop and evaluate AlzStack, a soft voting ensemble model for accurate Alzheimer's disease classification.
• To compare AlzStack's performance against traditional diagnostic approaches and other ensemble methods.

Main Methods:

• Utilized a comprehensive dataset of 2149 patients, including demographic, medical, lifestyle, and cognitive variables.
• Implemented a 5-fold cross-validation pipeline with randomized hyperparameter tuning and advanced resampling techniques (SMOTE, ADASYN, BorderlineSMOTE, SVMSMOTE) to address class imbalance.
• Employed Explainable Artificial Intelligence (XAI) methods for model interpretability.

Main Results:

• AlzStack achieved high performance metrics: 94.27% AUC, 93.26% accuracy, 89.17% precision, 92.11% recall, and 90.61% F1-score.
• The soft voting ensemble classifier outperformed stacking and hard voting ensembles.
• XAI methods identified key predictive features such as MMSE scores, functional measures, and behavioral markers.

Conclusions:

• AlzStack demonstrates robust predictive performance for early Alzheimer's disease detection.
• The model offers explainable insights into AD diagnosis, enhancing its utility as a healthcare decision-support tool.
• This AI approach provides a more efficient and accurate alternative to traditional AD diagnostic methods.