Improving Early Alzheimer’s Prediction with a Hybrid Ensemble Machine Learning Model

Abstract

It is essential to diagnose Alzheimer's disease (AD) at an early stage in order to provide rapid management and improve positive outcomes for patients. The purpose of this work is to demonstrate a mixed ensemble machine learning algorithm that can reliably predict the start of early-stage Alzheimer's disease by utilising clinical and neuropsychological datasets. The core of the model is a sequential pipeline that consists of data preparation, feature selection, and a classification architecture that is based on stacking on two levels. MICE, which stands for Multivariate Imputation by Chained Equations, is the method that we use in situations when there is a paucity of data. We scale the features by using the Z-score normalisation method. The Synthetic Minority Over-sampling Technique, more often referred to as SMOTE, is the technique that we have developed to address the issue of class imbalance. Through the process of reducing the number of dimensions in the model and pruning it, recursive feature elimination (RFE) is able to identify the most important characteristics and enhance the efficiency of the model. XGBoost, Support Vector Machine (SVM), and Random Forest (RF) are the three fundamental learners that are used in the suggested ensemble. These learners are combined with a variety of characteristics via the utilisation of a Logistic Regression (LR) meta-learner, which produces probabilities as its outputs. Using a stratified 10-fold cross validation process, we put the model through its paces and evaluate its accuracy, precision, recall, F1 score, and area under the curve (AUC-ROC). As part of our efforts to reduce the number of instances in which a false negative diagnosis is made, we are putting our whole of attention on memory. The hybrid framework that was provided is a viable hybrid strategy for early Alzheimer's disease detection, outperforming all individual models in terms of prediction performance and durability. This conclusion is based on the results of the experiments that were conducted.