An enhanced feature selection and cancer classification for microarray data using relaxed Lasso and support vector machine

Abstract Cancer is still the main cause of mortality for both men and women all around the world. In fact, about one in six deaths in the world is due to cancer, making it the most common cause of death globally. Lung and breast cancers had the highest mortality rates in men and women, respectively. Early detection of cancer is important to improve the chance of survival since early treatment can be provided for the patients who have this disease. The emergence of microarray technology has been applied to the medical field in terms of classification of cancer and other diseases. By using the microarray, the expression of hundreds to thousands of genes can be analyzed simultaneously. However, this microarray suffers from several problems such as high dimensionality, noise, and irrelevant genes. Thus, various feature selection methods have been developed intended to reduce the dimensionality of microarray as well as to select only the most relevant genes. In addition, it also difficult to select relevant features for classification from microarray gene expression data and successfully differentiate subgroups of cancer. For this study, we select three datasets of cancer microarray in the experiment. This chapter proposed relaxed Lasso and support vector machine (rL-SVM) for selecting features and classifying cancer. We gain classification accuracy through a 10-fold cross-validation for all datasets to compete with other existing methods. The performance of the classification algorithm will be evaluated by using the accuracy, area under the curve (AUC), and Kappa statistics. In this chapter, the experimental findings indicate that the method proposed has improved efficiency and achieves better accuracy for classification with fewer selected feature genes. rL-SVM can be used in large for classification of high dimension and small sample cancer data.