A correlation based multilayer perceptron algorithm for cancer classification with gene-expression dataset

Research on feature selection techniques for identifying informative genes from high dimensional microarray datasets has received considerable attention. Numerous researchers have proposed various optimized solutions to reduce noises, redundancy in dataset and to enhance the accuracy and generalization of the classification model by applying many computational tools. High-dimensional microarray gene expression dataset has limitations to many feature selection techniques with respect to generalization and effectiveness. A robust feature selection technique need to be designed which can remove irrelevant data, increase learning accuracy and improve comprehensibility of the experimental result. In this work, a novel correlation based feature selection algorithm using symmetrical uncertainty and multilayer perceptron algorithm is proposed. This method can identify the relevancy of the features to the class and also the redundancy considering all other relevant features of the dataset. It also evaluates the worth of a set attributes by measuring the symmetrical uncertainty with respect to another set of attributes. The effectiveness of the method is validated through various correlation based feature selection techniques using multi-category high-dimensional microarray datasets.