An Efficient Approach for Classification of Gene Expression Microarray Data

Abstract

Microarrays help in storing gene expression data from a cell. Each microarray describes features of each cell. The rows in microarray represent the samples and the columns represent the gene expression level of the cell. Microarray data is of high dimension due to which classification using conventional methods becomes tedious and inefficient. Therefore, reducing the dimension of long feature vector and extracting relevant features out of it becomes a very challenging task. This can be achieved using various techniques of feature extraction and/or feature selection. Design of an efficient classification model is another crucial task for any classification problem. In this paper, emphasis is given for significant feature extraction as well as efficient design of classifier. The task of microarray classification is done in two phases. In the first phase, a hybrid approach of Genetic Algorithm (GA) and Principal Component Analysis (PCA) is used for extracting relevant features. In the second phase, Probabilistic Neural Network (PNN) is used as the classifier and GA is implemented to optimize the topology of the PNN. The datasets used in the experiment are Colon Tumor, Diffuse Large B-Cell Lymphoma (DLBCL) and Leukaemia (ALL and AML). The proposed technique gave efficient results for the datasets used.