NoSHA: a Novel Stacked Hybrid Approach for Alzheimer's Disease Prediction

Alzheimer's disease (AD) is a kind of progressive neurodegenerative disorder affecting millions worldwide, and early diagnosis is a decisive factor in treatment outcomes. This paper proposes a machine-learning-based approach for predicting Alzheimer's disease using the OASIS dataset. Our approach combines the uses of the following eleven machine learning algorithms: Logistic Regression (LR), Decision Tree (DT), Support Vector Machine (SVM), Random Forest (RF), Naive Bayes (NB), Multilayer Perceptron (MLP), K-Nearest Neighbors (KNN), Linear Discriminant Analysis (LDA), XGBoost (XGB), AdaBoost (ADAB), and Gradient Boosting (GB). Feature selection techniques applied to enhance model performances in this study include Sequential Feature Selection (SFS), Recursive Feature Elimination (RFE), and Correlation Feature Selection (CFS). We developed an ensemble-based stacking classifier that combined several base models and yielded the best predictive performance. The CFS stacking model has 100% accuracy, precision, sensitivity, specificity, and F1-score, proving its predictive power. The stacking model using SFS has 97.73% accuracy, 100% precision, specificity, and sensitivity, and 94.74% F1 score. The stacking model using RFE has 96.59% accuracy, 86.96% precision, 100% sensitivity, 95.59% specificity, and 93.02% F1 score. Statistics show that feature selection improves Alzheimer's disease prediction, with CFS doing best. It proved highly effective for the successful prediction of Alzheimer's disease. It can favor early intervention strategies by diagnosing Alzheimer's disease much earlier than current techniques.