Evaluating the performance metrics of different machine learning classifiers by combined feature extraction method in Alzheimer's disease detection

Abstract

The rapid growth of machine learning technology has a great advantage in developing high performance neural network algorithms which can extract low-to-high-level features in human MRI images. Classification of clinical data for Alzheimer’s disease has always been challenging as currently there is no clinical test for Alzheimer’s disease. Doctors diagnose it by conducting assessments of patients’ cognitive decline. But it’s particularly difficult for them to identify mild cognitive impairment at an early stage when symptoms are less obvious. Also, it is difficult to predict whether patients will develop Alzheimer’s disease or not. The accurate diagnosis of Alzheimer's disease in the early stage is important to take preventive measures and to reduce the severity and progression before irreversible brain damages occur. The effectiveness of abnormality detection depends on the accuracy and robustness of the algorithm used. Different machine learning techniques with different levels of sensitivity, specificity, and accuracy have been developed. In this paper, performance measures of different classification algorithms like Support Vector Machine, k-Nearest-Neighbor, Discriminant Analysis Model Pseudo Linear, Discriminant Analysis Model Pseudo Quadra and Binary Decision Tree are compared and analyzed to obtain the most accurate classification method for the early prediction of Alzheimer’s disease.