From Complexity to Clarity: Improving Microarray Classification with Correlation-Based Feature Selection

Gene microarray classification is yet a difficult task because of the bigness of the data and limited number of samples available. Thus, the need for efficient selection of a subset of genes is necessary to cut down on computation costs and improve classification performance. Consistently, this study employs the Correlation-based Feature Selection (CFS) algorithm to identify a subset of informative genes, thereby decreasing data dimensions and isolating discriminative features. Thereafter, three classifiers, Decision Table, JRip and OneR were used to assess the classification performance. The strategy was implemented on eleven microarray samples such that the reduced samples were compared with the complete gene set results. The observed results lead to a conclusion that CFS efficiently eliminates irrelevant, redundant, and noisy features as well. This method showed great prediction opportunities and relevant gene differentiation for datasets. JRip performed best among the Decision Table and OneR by average accuracy in all mentioned datasets. However, this approach has many advantages and enhances the classification of several classes with large numbers of genes and high time complexity.