EARLY DETECTION OF ALZHEIMER’S DISEASE USING DATA MINING: COMPARISON OF ENSEMBLE FEATURE SELECTION APPROACHES

Abstract

Early Alzheimer's disease detection has become an important research area for many years. Various studies in the field of Alzheimer's disease detection have focused on applying individual feature selection methods. In addition to individual feature selection methods, the ensemble feature selection approach has become a creative field. It advocates the combination of the ranked features from various feature selection methods to obtain better results than the current approaches. Thus, this study aims to build a predictive model for early diagnosis of Alzheimer's disease using the ensemble feature selection approaches. Also, Alzheimer's disease dataset consists of three target classes: Normal (CN), Mild Cognitive Impairment (MCI) and Alzheimer's disease (AD). In this study, homogeneous and heterogeneous ensemble approaches have been applied in the feature selection process. Two feature subsets are created based on these ensemble feature selection approaches. A predictive model for early diagnosis of Alzheimer's disease has been build applying Random Forest, Artificial Neural Network, Logistic Regression, Support Vector Machine, and Naïve Bayes data mining algorithms. The predictive model uses the two feature subsets applying these algorithms separately. Then, the performance results are compared to determine which ensemble feature selection approach performs better than the other. This study revealed that better performance result is provided applying Random Forest algorithm with feature subset obtained using the heterogeneous ensemble feature selection approach (91%).