Optimal Gene Selection and Classification of Microarray Data Using Fuzzy Min-Max Neural Network with LASSO

Abstract

AbstractMicroarray gene expression data is a small sample high-dimensional dataset in which each sample is attributed with thousands of genes. The gene expression dataset is therefore very hard to classify because we have to consider thousands of genes for each sample while training the dataset. In this paper, we propose to classify the lung cancer microarray gene expression data using the Fuzzy Min-Max (FMM) classifier that is seldom used for high-dimensional datasets due to the large computational overhead. To improve the accuracy and speed of the FMM classifier, we use Least Absolute Shrinkage and Selection Operator (LASSO) to select the optimal gene subset for classification of lung cancer. We compare the classification performance of FMM-LASSO with that of support vector machine (SVM), Random Forest, K-nearest Neighbor (KNN), Naïve Bayes and Logistic Regression classifiers, with and without LASSO. The results prove that FMM-LASSO performs better as compared to other approaches.KeywordsMicroarray dataGene expressionLung CancerFuzzy Min-Max neural networkLASSO