Improvement of Machine Learning Models’ Performances based on Ensemble Learning for the detection of Alzheimer Disease

Abstract

Failure to early detection of Alzheimer’s disease (AD) can lead memory deterioration. Therefore, early detection of AD is essential affecting the points of the brain that control vital functions. Various early AD detection approaches have been employed using machine learning. In literature, most of the early detection of AD approaches has been developed using single machine learning methods. Due to the importance of early detection of AD, the goal of this study is to improve the classification performance of the previous studies for early detection of AD applying ensemble learning methods including bagging, boosting and stacking. ADNI clinical dataset was used in this study with three target classes: Normal (CN), Mild Cognitive Impairment (MCI) and Alzheimer’s disease (AD). The proposed ensemble learning methods provided better classification performance compared to single machine learning methods. Besides, the best classification performance from the ensemble methods is obtained through the boosting (AdaBoost) ensemble (92.7%). This study revealed that the classification rate increased up to between 3.2% and 7.2% compared to single based machine learning approaches through the AdaBoost ensemble method.