Modified PSO based feature selection for Microarray data classification

The main goal of successful Microarray data classification is to reduce the computational time while improving the classification accuracy. Though a large pool of techniques are already available, accurate classification of normal and malignant tissue cells is very challenging for the diagnosis of various types of cancers in humans. In this paper, Support Vector Machines (SVM), Naïve Bayesian and k-Nearest neighbor classifiers are used for classification of publicly available biomedical microarray datasets such as Prostate cancer, Leukemia and Colon tumor. To overcome the curse of dimensionality, Modified Particle Swarm Optimization (MPSO) is used to select the features from the datasets. A number of useful performance evaluation measures including classification accuracy, precision, recall, F-score as well as the area under the receiver operating characteristic curve are taken to evaluate the classifiers. After analyzing the experimental results, it is verified that SVM outperforms other classifiers and the performance is even improved a lot after feature selection.