An Extended Genetic Algorithm Based Gene Selection Framework for Cancer Diagnosis

Existing biomedical research shows that cancers can be diagnosed by effective feature selection and classification of DNA microarray in the molecular level. DNA microarrays are well-known as high-dimensional, possibly redundant, and small sample size data, which usually consist of many thousands of features. Therefore, feature selection methods are highly demanded to reduce the complexity for classification. Various approaches are proposed to select the most effective feature subsets for the classifiers to distinguish different cancer types. However, the optimal feature selection solutions are still difficult due to the trade-off between computational efficiency and classification accuracy. Our work shows that the genetic algorithm (GA) based hybrid feature selection methods balance the efficiency and accuracy. In this study, we propose a novel hybrid feature selection framework based on an extension of GA to obtain the optimal feature subset for gene expression data classification. Experimental results show that the proposed method is efficient, and extracts the minimal number of features with highest possible classification accuracy compared to existing methods.