Deep Learning approach for Microarray Alzheimer’s Data Classification

Abstract

Alzheimer’s disease (AD) has a quite complex genetic architecture and is an important progressive neurodegenerative disease. A major objective to the research of biomedical is to discover the genes that are risk then to explain the task of these genes in development of the disease. For such reason it is necessary to expand the set of genes which are associated to AD. Genes take central part in all biological processes. Microarray technology has provided genes with a huge number to measure many expression levels simultaneously. Microarray datasets characteristically have genes which are a huge number and samples with small size. This truth is described as a dimensionality curse that has a complex task. A promised method which named gene selection is solving such issue and has a major turn for creating an effective diagnosis of Alzheimer’s. In such study, methods of gene selection have been implemented, including Principle Component Analysis (PCA) and Singular Value Decomposition (SVD). Such methods have the ability to reduce the number of insignificant and genes that are redundant in the original datasets. After that, deep learning (DL) via Convolutional Neural Network (CNN) serves as a classifier to predict AD. CNN which consists of six-layer having various parameters for the dataset has been used. The empirical results are showed with AD dataset that PCA-CNN model achieves 97.24% accuracy and loss 0.4614 while SVD-CNN model achieves 98.99% accuracy and loss 0.2588. Thus, the proposed system is suitable for decreasing the genes dimensions by means of selecting subset of informative gene and enhance the classification accuracy.