Research On Application Of Customer Segmentation Model Based On Consumption Behavior

The tremendous competition our financial industry is facing is forcing banks to adopt a client-centered management model after our country's joining to WTO. In a client-centered management style, the most important and urgent is to classify the customers properly. Grouping customer into categories gives us better understanding of customers and provides personalized service.There has been some research abroad on consumer behavior applied in the field of consumer segmentation, like RFM model, Customer Value Matrix (CVM) model, etc. Most of the segmentation criterions in current practice are produced by customer geographic distribution and their characteristics, but seldom can we find those criterions based on customer behavior. These segmentation model are classified only by one side of the customer's behavior. In this paper, we present a new clustering approach to group customers and define consumption expenditure and consumption fluctuation rate as segmentation variable to determine the subdivided variables to overcome the shortcomings of the existing subdivision.This paper aims at establishing a customer segmentation model based on consumption behavior. The model can guide the banks to detect targeted marketing and fraud risk. Firstly, we use improved K-MEANS algorithm to construct the model, from which we can not only get characteristics of consumer behavior patterns, but also find interested customer groups through identifying characteristic patterns. Secondly, we use DBSCAN algorithm to construct the model. Through comparing the two results, we find that DBSCAN excludes the fact that an isolated point has an influence of clustering in the process of clustering, giving the customer types closer to reality.Finally, we try to apply this model to detect credit card fraud. Experiments show, the correct detection rate used by BP algorithm after customer segmentation has increased slightly, while error detection rate has decreased. This proves that the segmentation method which we have used is effective.