Quadratic kernel-free least squares support vector machine for target diseases classification

Support vector machines (SVMs) have been proved effective and promising techniques for classification problem. Recently, SVMs have been successfully applied to target diseases classification and prediction by using real-world data. In this paper, we propose a new quadratic kernel-free least squares support vector machine (QLSSVM) for binary classification problem. The model of QLSSVM is a convex quadratic programming problem with an advantage of kernel-free, compared with the existed least squares SVM. By using consensus technique, the decision variables of QLSSVM are split into local variable and global variable. Then the QLSSVM is converted into the consensus QLSSVM and solved by alternating direction method of multipliers with a Gaussian back substitution. Finally, our QLSSVM is illustrated in terms of numerical tests based on two types of training data sets. The first numerical test is implemented based on artificial data to certify the performance of our QLSSVM. To apply our QLSSVM to disease classification, the second one is implemented based on diseases data set from University of California, Irvine, Machine Learning Repository to demonstrates that our model has higher classification accuracy compared with several existed methods. In particularly, our numerical example is implemented based on a special heart disease data set provided by Hungarian heart disease database to illustrates the effectiveness of our QLSSVM for a particular disease diagnosis.