Gene selection by Mutual Nearest Neighbor approach

Gene expression data suffer from the curse of dimensionality due to the presence of several thousands of genes (features) but a small number of samples. This problem of large feature space is addressed by feature selection algorithms which aim at finding a comparatively small set of significant features by removing the redundant and irrelevant features thereby increasing the performance (e.g., higher accuracy for classification), decreasing the computational cost and improving the model interpretability and comprehending the results in an better way. In this paper, we explore the possible application of Mutual Nearest Neighbor (MNN) and Mean Test approaches to select significant genes from high dimensional gene expression data and compare their performances with three other well known algorithms. kNN classifier is used to measure the performances of these algorithms and the results are illustrated.