Font Size: a A A

Three Classification Algorithms Based On Nonparallel Support Vector Machines

Posted on:2020-04-26Degree:MasterType:Thesis
Country:ChinaCandidate:L N LiuFull Text:PDF
GTID:2428330572489719Subject:Operational Research and Cybernetics
Abstract/Summary:PDF Full Text Request
Support vector machine attracts more and more attention with its excellent learning abil-ity.Traditional support vector machine,such as standard support vector machine,?-support vector machine,least squares support vector machine and so on,all are based on the principle of maximum margin to find a single decision separating hyperplane,which limits its general-ization.while the nonparallel support vector machine is different,it constructs corresponding support hyperplanes for each sample,obtaining a pair of nonparallel decision separating hyper-planes,thus applicable to wider problems.The existing nonparallel support vector machines have their own advantages,but there are still some shortcomings,resulting in low classification accuracy for some data sets.Therefore,this thesis starts from improving the classification ac-curacy and versatility of the algorithm.Some nonparallel support vector machines are studied,and the following three new classification algorithms are proposed.1.For existing ?-nonparallel support vector machines and L1-nonparallel support vector machines(termed as L1-NPSVM),each of which has its own unique advantages and disad-vantages.In thesis,we propose the L1-?-nonparallel support vector machine(termed as L1-?-NPSVM),which can naturally inherit their unique advantages while overcoming its short-comings,guaranteeing the uniqueness of the decision function and dealing with the unbalance problem.2.To further improve the applicability of L1-NPSVM,by replacing the e-insensitive loss function in L1-NPSVM with the soft-margin e-band insensitive loss function and minimiz-ing the width of the ?-band,the soft-margin ? nonparallel support vector machine(termed as INPSVM)is obtained.The model can degenerate into the traditional support vector machine,at the same time,it is applicable to a wider range of nonparallel structure data.3.For the previously proposed INPSVM,its parameter C is difficult to take values.By introducing The parameter v replaces the parameter C of INPSVM,this thesis proposes the?-soft-margin ? nonparallel support vector machine(termed as ?-INPSVM).?-INPSVM not only reduces the difficulty of selecting parameters,but also has a stronger promotion.
Keywords/Search Tags:Support vector machines, classification problems, loss function, kernel function, convex quadratic programming problem
PDF Full Text Request
Related items