Cancer Classification from Gene Expression data using Fuzzy-Rough techniques An Empirical Study

Cancer classification from gene expression data is one of the most challenging research areas in the field of computation biology, bioinformatics and machine learning as the number of clinically labeled samples are very few compared to number of genes present. Also the cancer subtype classes are often highly overlapping, imprecise, and indiscernible in nature. Various machine techniques have been developed and applied on gene expression data for cancer sample classification. Here in this article, an empirical study of cancer classification from microarray gene expression data is performed using fuzzyrough nearest neighbour techniques where performance of four different types of classifiers viz., Fuzzy nearest neighbour, Fuzzy-rough nearest neighbour, Vaguely quantified fuzzy-rough nearest neighbour and Ordered weighted average based fuzzyrough nearest neighbor are investigated. The experiments are carried out on eight publicly available real life microarray gene expression cancer datasets. To assess the results of the classifiers percentage accuracy, precision, recall, macro averaged F1 measure, micro averaged F1 measure and kappa are used. The comparative study of the investigated methods is also done using paired t-test. Fuzzy-rough nearest neighbour method is found to be better for most of the data sets for cancer classification. Keywords—Cancer Classification, Fuzzy-Rough set, Vaguely Quantified, Ordered Weighted Average, Microarray Gene Expression data.