基于OSA算法和GMDH网络集成的电子商务客户流失预测

E-Business Customer Churn Prediction Based on Integration of Objective System Analysis and Group Method of Data Handling Network

电子商务客户流失预测是一种典型的高维、非线性、数据不平衡问题,传统的方法已很难提高其预测精度。本文将自组织数据挖掘方法(SODM)引入电子商务客户流失预测,提出一种基于客观系统分析(OSA)和数据分组处理(GMDH)网络集成的电子商务客户流失预测模型。首先利用OSA算法自动选择出重要的电子商务客户流失关键属性,然后将训练样本送入GMDH网络进行学习与训练,进而对测试样本客户流失状态进行预测。为了提高预测精度,本文还利用向上采样法进行数据平衡化,使得流失类和非流失类客户数量大致相等。应用该模型对某网上商场客户流失状态进行预测,并将预测结果与神经网络、SVM等方法得到的结果进行了比较,验证了该模型的有效性及实用性。

Facing with the high dimensional,nonlinear and unbalanced data problems of churn prediction of E-business customers,it is difficult to improve the accuracy of churn prediction of E-business customers by applying traditional methods.Hence an integration model for churn prediction of E-business customers based on objective system analysis（OSA） and group method of data handling（GMDH）,two important self-organized data mining（SODM） algorithms,is presented in this paper.Firstly,the key attributes are automatically selected using OSA algorithm.Then GMDH network is trained with training samples,which is used to identify customer churn status of testing samples.Up-sampling metod is also used in this paper to balance the churn-customer data and unchurn-customer data to improve the forecasting accuracy.This proposed approach is applied for chum prediction of an online shop,which proves that compared with some common approaches such as artificial neural networks and support vector machines,more accuracy forecasted results can be obtained.