Research On Structural Fuzzy Twin Support Vector Machine And Its Solving Method

The twin bounded support vector machine finds non-parallel hyperplanes by solving two smaller-scale quadratic programming problems,which further improves the classification performance of the twin support vector machine.However,this method mainly uses the Hinge loss function when constructing the model,which makes the twin bounded support vector machine more sensitive to noise and unstable to resampling;in addition,the problem needs to be converted from the original space to the dual space which reduce the time performance of the algorithm to solve the model.In order to handle the shortcomings of the twin bounded support vector machine better,the structural information and fuzzy membership are introduced into the twin bounded support vector machine in this paper,and the corresponding twin bounded support vector machine models with different loss functions are constructed.Using smoothing technology,the solving method of structural fuzzy twin bounded support vector machine model is studied.On this basis,different smooth twin bounded support vector machines are proposed.The main work is as follows:1.Research on twin bounded support vector machine model based on Pinball loss and its solving method.Aiming at the noise sensitivity and resampling instability problems in the Hinge loss function,the Pinball loss function is introduced into the twin bounded support vector machine,and the twin bounded support vector machine model based on the Pinball loss is constructed.The non-differentiable problem of the Pinball loss function at the zero point is solved by constructing a smooth approximation function,and a smooth twin bounded support vector machine model based on the Pinball loss is proposed.The Newton-Armijo method is used to solve the problem iteratively in the original space.On this basis,a algorithm of smoothing twin bounded support vector machine based on the Pinball loss is proposed,and the proof of the convergence of the algorithm is given theoretically.2.Research on structural fuzzy twin bounded support vector machine model based on Pinball loss and its solving method.In view of the prior knowledge and distribution information between the samples in the data set,and each sample plays a different role in the process of generating the classification hyperplane,the structural information of the data set and the fuzzy membership of the sample are introduced into the twin bounded support vector machine model of the Pinball loss,the corresponding smooth structural fuzzy twin bounded support vector machine model is obtained,and the smooth structural fuzzy twin bounded support vector machine algorithm based on Pinball loss is proposed.In addition,since there is a defect that reduce the sparsity of the solution in the model solving,the structural fuzzy twin bounded support vector machine model based on the ?-Pinball loss is further studied,and the smooth approximation function of the ?-Pinball loss function is constructed.Using the unconstrained smoothing processing,a smooth structural fuzzy twin bounded support vector machine algorithm based on ?-Pinball loss is proposed.3.Research on the structural fuzzy twin bounded support vector machine model and its solving method based on L1 loss.Aiming at the structural fuzzy twin bounded support vector machine model of L1 loss,through smooth and unconstrained processing of the objective function,a smooth structural fuzzy twin bounded support vector machine model of L1 loss is obtained.By solving this model,an algorithm of structural fuzzy twin bounded support vector machine based on L1 is proposed.4.In the experiments,to verify the proposed model,some representative methods are compared with the proposed method on the selected UCI data set,artificial data set,and NDC data set.And the experimental results show that the proposed solving algorithm of smooth model saves running consumption,and can obtain better classification accuracy,thus the effectiveness of the proposed algorithm is verified.