Research On Self-tuning Fuzzy Support Vector Machine Algorithms For Shifting Class Center

As an excellent Machine learning algorithm,Support Vector Machine(SVM)has been widely used in industrial production and daily life.However,SVM cannot process Fuzzy information and is sensitive to noise.Therefore,scholars hope to solve the above problems by introducing Fuzzy Support Vector Machine(FSVM).The improvement of FSVM on SVM is to obtain the Membership Function(MBSF)through the analysis of the characteristics of the sample data set and to give each sample a Membership(MBS)through MBSF.Therefore,the core of FSVM algorithm is how to construct a good MBSF.According to the different construction of MBSF,scholars proposed that the model can be divided into two types:FSVM designed by MBSF based on Sample and Class Center Distance(DSC)and FSVM designed by MBSF based on Sample and Class Center,Hyperplane Distance(DSCH).The most original algorithm proposed by FSVM based on DSC to design MBS has been widely applied due to its simple and easy operation.But this design of MBSF will inevitably lead to different membership of support vectors,which will affect the classification accuracy.Although the FSVM algorithm designed based on DSCH can solve the above problems to some extent,the time complexity of the algorithm based on DSCH is relatively high because DSCH needs to be calculated,and the forced use of DSCH reflects the poor universality of MBS algorithm.Based on some improved algorithms proposed by scholars based on two types of traditional FSVM models.The paper proposes to introduce adjustment factors after analyzing Potential Support Vector Samples(PSVS),making class center can be adjusted along the direction of classification plane.In this way,the two self-tuning FSVM algorithms in this paper are obtained by using the traditional two-class FSVM model on the basis of the adjusted class center.In this way,the improved algorithm can integrate the advantages of the two traditional models,and will not bring a large time cost At the same time,the traditional FSVM is improved in experimental section:in this article,through comprehensive nuclear parameters and punish coefficient of two variables,to the main nuclear parameter variables as independent variables,to punish coefficient to get the average classification accuracy of secondary variables as the dependent variable,through two-dimensional curve changes to compare the differences between algorithm,in this way,the results more convincing.In the experimental part,this paper obtained the corresponding optimal deviation proportional coeffcient through the experimental analysis of two types of improved models,and then compared the classification accuracy and classification time of the two types of improved FSVM algorithm and other FSVM algorithms.Experiments show that the two self-tuning fuzzy support vectors proposed in this paper are optimized in stability and classification accuracy under appropriate time complexity.At the same time,the application scenarios of the two improved models are obtained by analyzing the experimental results.Figure 27 table 5 reference 58...