Research On Algorithms Of Quadratic Surface Support Vector Machine

In this paper,we study the models and algorithms of quadratic surface support vector machine.The performance of a traditional support vector machine depends heavily on the selected parameters of the kernel function.And its classification accuracy is easily affected by outliers and noise.Quadratic surface is found to separate the training sample set directly without using any kernel function.And then,loss function can measure the difference between the real value of sample and the value predicted by model well.We introduce a variant of logistic regression loss function into quadratic surface support vector machine.Variant logistic regression loss function quadratic surface support vector machine(LQSSVM)is proposed then.Results of numerical tests on some actual classifying data sets manifest that the model has a good performance in classification.Moreover,the purposed model shows more accurate classification than soft quadratic surface support vector machine(SQSSVM)and soft support vector machine with Gauss kernel function(SKSVM).