A Machine Learning-Based Approach for Predicting Alzheimer's Diseases Using Data Classification and Explainable Artificial Intelligence Techniques

Globally, healthcare systems encounter tremendous obstacles as a result of the progressive

neurological disease called Alzheimer's. To effectively intervene and control Alzheimer's disease

(AD), the earliest and most accurate diagnosis is necessary. However, traditional diagnostic methods

are often expensive, take a long time, and do not provide the accuracy needed for early diagnosis.

This study addresses these limitations by proposing a machine learning-based (ML-based) approach

for predicting AD using advanced data classification methods and an explainable artificial

intelligence (AI) approach. Three distinct methods were utilized to carry out the feature selection

procedure: chi-square, mutual information, and analysis of variance (ANOVA). We identified the

analysis's most relevant elements by utilizing each technique. We found the best algorithm for

predicting the early signs of AD by testing seven different ML methods: logistic regression,

AdaBoost, random forest, support vector machine, decision tree, XGBoost, and K-nearest neighbors.

We employed the SMOTE method to rectify the data imbalance. To test the proposed method, we

employed both a publicly available and a private dataset. We applied multiple cross-validation

approaches to provide a strong performance evaluation. The results of the experiments illustrated

that, out of all the models tested, the XGBoost classifier performed the best. Using the combined

dataset, the XGBoost classifier had 97.32% accuracy, 96.56% precision, 97.00% specificity, 97.68%

sensitivity, 98.43% AUC, and 97.12% F1-score. Using the public dataset, XGBoost achieved

97.23% accuracy, 96.14% precision, 96.52% specificity, 98.03% sensitivity, 98.30% AUC, and

97.07% F1-score. Furthermore, XGBoost did exceptionally well on the private dataset with 95.83%

accuracy, 93.94% sensitivity, 96.88% precision, 97.44% specificity, 98.52% AUC, and 95.38% F1-

score. Understanding the model's findings and decision-making process can be enhanced with the

help of an explicable AI framework that was developed using SHAP methods. The proposed

approach shows enormous potential as a healthcare solution that reduces healthcare costs and

improves efficiency in AD’s diagnosis. Patients benefit from improved diagnostic tools for AD

brought about by this study's combination of powerful ML models with explainable AI.