The Research Of Customer Churn Prediction Based On Data Mining Methods

With the increasingly keen industry competition caused by the globalization of economy, enterprises in Information Age are compelled to capture opportunities and build up their core competition ability by utilizing knowledge concealed in large amount of data.For most enterprises, customers are the key success factor and the most important source of profit. Customer Relationship Management (CRM) based on data mining techniques provides a quantitative criterion in business management and decision-making. CRM helps enterprises utilize their limited resources more effectively so as to broaden their profit development space.It is a tough problem for all enterprises to predict and control customer churn. They have suffered heavy losses caused by the frequently occurred customer churn that prolongs their cost recover cycle. Research routine of decreasing customer churn is to provide customized service, or analyze customers' satisfaction and loyalty. The effectiveness of these methods is hard to be verified. Furthermore, they could not solve the problem essentially.In this thesis, data analysis techniques are merged into the research of customer churn. Solutions of existing problems involved are proposed in detail, including customer profiling, attribute reducing, customer churn model building, analysis of churn causation, churn prediction and controlling strategies. Among those we focus on the customer churn model building problem.Customer churn model, consists of basic model and behavioral model, is built based on analysis of the churned customers' basic data and behavioral history. The setof key attributes that describe churned customers' basic character are founded by applying association rule mining to their basic profiling data. After adjusting and revising algorithms proposed by precedent researchers, we develop a Constrained Adaptive Association Rule Mining (CAARM) algorithm that can find rule with specific head, and it does not require a given minimum support. Sequence pattern discovery method is used to build behavioral churn model, by which we can distinguish the typical behavioral sequences of the churned customers and predict present customers' churn tendency.Based on the superficially discussed issue of customer lifetime value, we suggest that customer subgroups with higher lifetime value should be selected as target of specific market strategy.Finally, there are central algorithms described in detail and the analysis results of them as well.