Bounded Sum Of Margins Support Vector Ordinal Regression

Ordinal regression,also known as ranking learning,is a supervised learning problem between classification and regression in the field of machine learning.The goal of ordinal regression is to learn a multi-classifier from the samples with ordinal labels to predict the labels of new samples.Many practical problems attach great importance to the treatment of the ordinal relationship of labels.Most of these problems can be modeled as ordinal regression problems.In fact,in the fields where human needs,behaviors and preferences play an important role,ordinal regression has been widely used,such as medical research,credit rating,text classification,face recognition,social science and so on.Support vector machine(SVM)is a kind of model with good generalization performance in machine learning.It can establish effective multi-classification algorithms to deal with ordinal regression problems.Therefore,a variety of support vector ordinal regression(SVOR)models have been formed.Based on the maximum margin criterion of SVM,two strategies which are used to solve the SVOR problems are formed: fixed margin strategy and sum of margins strategy.Multiple single ranking SVM(MSRSVM)is a model based on the sum of margins strategy,which can make better use of the sorting information of samples.In order to fully consider the overall distribution of data,we add boundaries at both sides of the MSRSVM model to prevent some points deviating from the overall direction of the data set from being recognized as support vectors,and then control the direction of the decision hyperplanes close to the overall direction of the data set.Finally,the bounded sum of margins support vector ordinal regression(BSMSVOR)model comes into form.BSMSVOR model can not only make full use of the sorting information of samples,but also consider the overall distribution of data sets,so it can be better applied to practical problems.Numerical experiments on some ordinal regression data sets verify the effectiveness and feasibility of BSMSVOR model.Finally,we collect some data about Shanghai Stock Exchange 50 index and turn the question of whether the stock will rise or fall into an ordinal regression problem.We verify the good classification effect of BSMSVOR model again and successfully apply BSMSVOR model to the prediction of stock’s rising or falling.