Improving Alzheimer's Disease Prediction Accuracy using Feature Selection, K Fold Cross Validation, and KNN Imputer Techniques

Alzheimer's disease (AD) is a progressive neurodegenerative disorder characterized by cognitive decline and memory loss; it accounts for 60–70% of dementia cases. Early diagnosis remains challenging due to the subtlety of its symptoms. This study explores the effectiveness of ensemble methods, feature selection techniques, and imputation strategies in enhancing the accuracy of AD diagnosis. We applied an ensemble method with Chi-Square feature selection, achieving a high accuracy of 95.733% with 7 optimal features. The combination of classifiers, including Gradient Boosting (GB), Support Vector Machine (SVM), and Logistic Regression (LR), contributed to the high performance. Additionally, the use of KNN Imputer and K-Fold Cross Validation significantly improved accuracy, regardless of whether feature selection was employed. Notably, feature selection slightly reduced model complexity but resulted in a marginal decrease in accuracy. The study highlights the importance of these methods in achieving reliable AD predictions, though dataset dependency and potential biases from methodological choices are acknowledged. Future work may involve exploring alternative classifiers and validating findings across diverse datasets to enhance generalizability and address these limitations.