A Deep Learning Model for Prediction Alzheimer's disease Based on Microarray Gene Expression Data

Abstract

Alzheimer's disease (AD) is a major productive neurological illness with complicated genetic architecture. One of the main aims of biomedical research is to identify risk genes and then explain how these genes contribute to disease development. As a result, it is required to increase the list of genes linked to Alzheimer's disease. Genes play a crucial role in every biological activity. Microarray technology has given genes access to a large number of genes, allowing them to evaluate several levels of expression at the same time. Microarray datasets are categorized by a huge number of genes and small sample sizes. This reality is referred to as a multidimensional curse with a difficult task. A promising technology known as gene selection is addressing this issue and has the potential to revolutionize Alzheimer's disease diagnosis. In this work, gene selection approaches such as Singular Value Decomposition (SVD) and Principle Component Analysis (PCA) were used. Techniques can help to minimize an amount from trivial and redundant gene in the unique datasets. Then, using the Convolutional Neural Network (CNN) as a classifier, deep learning (DL) is used to predict AD. The dataset was processed using a CNN with seven layers and varied settings. With the AD dataset, the empirical findings reveal that the PCA-CNN model has 96.60 percent accuracy and a loss of 0.3503, while the SVD-CNN model has 97.08% accuracy and a loss of 0.2466. As a result, the suggested approach is suited for reducing gene dimensions and improving classification accuracy by choosing a subset of relevant genes.