Comparison of diagnostics set and feature selection for breast cancer classification based on microRNA expression

MicroRNA(miRNA) expression that have great potential serving as cancer biomarkers and therapeutic targets generally has a very large number and has brought great challenge for identification of the most feature sets. In this paper, combinatorial miRNA biomarkers from the diagnostic set and feature selection are comprised for breast cancer classification using Naive Bayes and backpropagation. The diagnostic set of miRNA are provided from recent bioinformatics and medical research results. Moreover, greedy stepwise using Naive Bayes and Multi Layer Perceptron method are utilized for feature selection in order to reduce number of miRNA set from 1881 features. MiRNA expression in Cancer and normal breast cells are examined to study this comparison. The classification performance of input sets were implemented and studied thoroughly in terms of Sensitivity, Specificity, Classification Accuracy and ROC value. Based on experimental results, this study obtained recommended features for cancer classification with less number than diagnostic sets and in this study, the essential features for cancer analysis are discovered as new essential biomarkers.