A Robust Ensemble Framework for Alzheimer’s Disease Prediction Using Feature Engineering and Optimized Voting Classifier

The progressive neurodegenerative condition known as Alzheimer’s disease (AD) is marked by memory loss and cognitive decline, making early detection and treatment extremely difficult. In order to enhance patient outcomes and facilitate treatments that can slow the disease’s course, it is critical to identify Alzheimer’s disease (AD) as soon as possible. To achieve this goal, this work presents a reliable strategy for identifying Alzheimer’s disease that combines feature engineering, optimization, and an ensemble approach. The dataset, which was acquired via a publicly available dataset from Kaggle, has 2149 patient records with 34 different features that cover a broad range of clinical, lifestyle, and demographic characteristics. Using Out-of-Fold (OOF) predictions via 5-fold stratified CV, a stacked ensemble model with CatBoost, XGBoost, LightGBM, and Random Forest is created as base learners and logistic regression as the meta-model. Both Attention-based Deep Non-negative Matrix Factorization (Attn-DNMF) and manual feature engineering were employed as feature extraction techniques. The strength of domain-driven feature creation was demonstrated by the 97% accuracy of the manual features and the 89% accuracy of the NMF-based method. With 97% accuracy, the model provides a workable way to detect Alzheimer’s at an early stage, which could improve patient care, reduce carer stress, and aid in healthcare decision-making.