Study On Retail Customer Behaviour Based On Support Vector Machine

As rapid economic development recently, the consumption has beennoticeably increased, which brought tremendous pressure to the storageof enterprise CRM system, especially for large-scale retail enterprises.The traditional database technology has been unable to meet the demandof modern analytical capacity faced to an ever-expanding variety of staticand dynamic mass of data resources. So it has become the key point forfuture research on how to find the rules and valuable knowledge hiddenbehind the massive data. With the constant improvement and innovationin data mining, at present, to some extant it has been achieved in somepractical applications, such as Decision-making tree, Bayesian and Neuralnetwork, etc. This paper proposed the Support Vector Machine (SVM)used in analysis of consumer behavior (CB) in large retail enterprise.SVM is a new pattern recognition method based on statistical learningtheory, which displays many unique advantages in resolving the limitedsamples, nonlinear and high-dimensional pattern recognition.This paper mainly focus on the following three aspects: Firstly,described the concept and meaning of retail CB. Then prove that online CB is a one of the important research attention for enterprise in futureafter off-line and on-line analysis of CB. Second, data mining techniquessupport for CB analysis are introduced and mainly study the SVM theory;a deep analysis was given to the mechanism, content and algorithm ofSVM combined with the statistical theories, optimal theory as well asversus theory. The character of SV and kernel functions is also listed.Third, construct an analysis model based on analysis of CB-DCSS model.The model can be used to deal with high dimensional-linear-inseparableproblems; the cloud-like process with mapping mechanisms and uncertainreasoning can reflect the relation of multi-attributes. Application ofcustomer consumption data algorithms to generate the classificationresults within the improved SMO combined the probability distributionfunction. Finally, it presents a specific example; DCSS-SMO algorithm,compared with other ones, is proved more efficient and accurate. Whichcan also applied to linear but not unclassified situation in consumerclassification. In this article, it provides a very effective tool after SVMintroduced into retail customer classification for enterprises to find thekey customers.