The Comparation And Research On Three Support Vector Classifiers Named As C-SVC??-SVC And LSSVC

Support Vector Classification(SVC)is a new machine learning algorithm proposed by Vapnik in the 1990 s.The most common and typical methods are C-Support Vector Classification(C-SVC),?-Support Vector Classification(?-SVC)and Least Squares Support Vector Classification(LSSVC).We need to compare the three SVM classifiers to help us understand the usage and characteristics of SVM more deeply,so as to better apply it to practical classification problems.This paper first explores the similarities of the three SVM classifiers: when the parameters of C-SVC and ?-SVC satisfy a certain relationship,the optimal partitioning hyperplane of C-SVC and SVC is the same,and the classification results of the two algorithms are consistent;When the soft interval loss function in C-SVC is modified to a quadratic soft interval loss function to obtain a two-norm support vector classification(L2SVC),the optimal partitioning hyperplane of L2 SVC and LSSVC is the same at that time,and the classification results of the two algorithms are consistent.Then the differences of the three support vector classifiers are explored:(1)Comparison of algorithm performance: when the number of training samples increases,the accuracy,precision,recall and F1-measure of the three classification algorithms will be improved;For binary classification problems,the accuracy,precision,recall and F1-measure obtained by ?-SVC algorithm are the best.For unbalanced data sets,the recall of LSSVC algorithm is poor.(2)Comparison of extendibility: When the sample set is small,the corresponding operation time of the three algorithms is relatively short.When the sample set is large,the operation time of the three algorithms is significantly increased,but the speed of LSSVC is obviously better than that of C-SVC and ?-SVC,and LSSVC has better extendibility.(3)Comparison of anti-interference performance of the algorithm: for the input of outliers with large deviation,the C-SVC algorithm has the strongest anti-interference performance and will not produce any impact;?-SVC algorithm is also relatively good in anti-interference,that is,the influence is not too big,while LSSVC algorithm has the worst anti-interference,but when the sample is large enough,a few outliers will not have a great impact on LSSVC algorithm;For the addition of intermediate fuzzy points,LSSVC algorithm and ?-SVC algorithm have a certain anti-interference,C-SVC algorithm has the worst anti-interference,the optimal partition hyperplane deviation is large,even if the increase of samples will still have a greater impact.