A Muitiple Attribute Decision Making Method Based On Support Vector Machine

Muitiple Attribute Decision Making(MADM)is also called multiple objective decision with finite alternatives,considering multiple solutions and each solution is described by multiple attributes.This kind of problem is common in many areas,so the research has a profound theoretical significance and wide practical application background.In the MADM problem,it is expected to give decision-makers alternative solutions objectively and fairly.Based on this consideration,this thesis proposes a computing mechanism which uses principle of Support Vector Machine(SVM),and makes regression analysis of the problems and gives decision-making model.SVM is a new machine learning technique based on the statistical learning theory and it has theoretical properties as well as the considerable performance of learning.Today,the statistical learning theory is in the transition period from theory to practice,the algorithms of SVM must be improved for the need of the practical application.Especially,SVM shows good performance in small sample and non-linear problems.So,this thesis makes use of the fitness of SVM to MADM,proposes a MADM method based on SVM.Firstly,this thesis discusses the selection of SVM model parameters.To make the parameters selection more faster,we introduce Particle Swarm Optimization(PSO).Then apply this model to a class of MADM problems,making regression and giving decision solutions.The contrast experiment shows performance of the optimization model.Secondly,an experimental analysis is done on journal evaluation.Because of the strong correlation between the attributes of journal evaluation,some regression methods are restricted.But the support vector regression method can be used without considering the relationship between attributes.At last,we do some further study on more complicated MADM problems with more attributes.Principal Component Analysis(PCA)is introduced to simplify this type of problem.Removing redundant and unimportant attributes at the extreme,then use SVM model to do regression fitting.It can improve the operation efficiency greatly and make decision easily.