Abstract: The improvement of microarray technology has facilitated scientists in the direction ofmonitor mRNA expression level of up toward tens of thousands of genes concurrently, inside just one single experiment on a solid surface, made of any glass or silicon. The huge amounts of dataset samples are then exploitedthrough biologists in the direction ofclassify cancer-related genes by means of comparing their expression values in healthy and cancerous tissues. Though, microarray dataset samples generally have thousands of genes or features and simply dozens of samples. Handling of huge amounts of dataset samples are most important challenging task in microarray dataset samples classification .So feature selection or reduction is akey important step in classification of microarray dataset samples with many thousands of features. As a wrapper method, Support Vector Machine-Recursive Feature Elimination (SVM-RFE) is one of the most important feature selection techniques. Though SVM-RFE is able toeliminate irrelevant features successfully, it mightn’t deal with the majority of the irrelevant features. In addition SVM-RFE also performs better than other methods by considering threshold value?. They mightbe several numbers of ways for choosing of this parameter thusresults higher performance or improved performance. But sometimes they decreases results of the classifier, to solve this problem Artificial Bee Colony (ABC) is introduced in this paper for automatic selection of threshold ? toward enhance classification results in terms of accuracy, precision and recall.This ABC algorithm is performed based on the behavior of colony of three bees: employee, onlooker and scout bees. Then, Deep Feed Forward Neural Network- Recursive Feature Elimination (DFFNN-RFE) is introduced for classification of microarray dataset samples. In DFFNN classifier is the performed based on the calculation of probability distributions that stochastically adjust small parts of the input data for the period of training phase. In addition the proposed work correlation coefficient is introduced in order to filteringof highly redundant features before performing DFFNN-RFE classifier. The experimentation results conclude that the proposed DFFNN-RFE classifier performs better when compared to a nonlinear SVM-RFE, classifier in terms of classification accuracy, precision and recall to cancer dataset.
Keywords: microarray technology, Support Vector Machine(SVM), Recursive Feature Elimination (RFE), Deep Feed Forward Neural Network(DFFNN), Artificial Bee Colony (ABC), feature selection, classification.