Research On SK Algorithm Of Support Vector Machines And Modeling Analysis In Customer Relationship Management

Customer relationship analysis is a very important part in customer relationship management(CRM). At the same time, data mining(DM) is a useful tool for data processing and knowledge discovery. In this paper, DM models and related algorithms are studied carefully to complete the tasks of customer churn prediction and customer lifetime value analysis which are the main issues in customer relationship analysis.The main work and results in this paper include:1. We give a short review on the customer churn analysis and related data mining methodology applications in customer churn. We also discuss related topics in customer churn analysis.2. We demonstrate the characters of customer churn prediction and formulate its models by means of support vector machines(SVM) algorithms. We introduce the Schlesinger-Kozinec(SK) algorithm to solve SVM model for large training problems and improve the SK algorithm in two forms: Forward-SK algorithm and Backward-SK algorithm. The Forward-SK resulting algorithm is simple to implement and as the theory analysis and experiments show competitive to the SK algorithm. While Backward-SK is a more general model and can be regarded as the special SK algorithm or Soft-SK algorithm under some conditions.3. We investigate the customer lifetime value (CLV) in customer churn and formulate CLV as a Markov Chain (MC) model. We make a study to the two kinds of MC models—with and without the state keeping probability (SKP) items. By means of Pfeifer's model, in infinite horizon purchasing case, we establish the two kinds of MC equations for the two cases, and get their expected net present value (ENPV) solutions by each computed analytic inverse matrix, then make a character analysis to each case; Then, we propose corresponding CLV (The total expected net present value in customer life cycle) results, and compare the two kinds of CLVs and make an analysis to the result. The study shows: the customer's CLV varies with its state keeping probability, and the scenario with SKP is better than that of without SKP. The model result is consistent with what we usually see in practice sales.The DM models and related models studied in this paper can be used in other areas of CRM.