对电信客户流失数据分别构建贝叶斯模型和SVM模型,进行电信客户流失的可能性预测.在实验过程中改变数据量和特征字段,借助clementine 12.0的可视化实验平台直观、有效地观察5种模型的预测结果,并对贝叶斯和SVM的5种模型进行比较,得出结论:在属性值较多的情况下,采用贝叶斯Markov-FS模型;在属性值较少且与预测结果高度相关的情况下,SVM中多项式核函数模型预测结果的正确率和稳定性都比较好.%Based on the telecom customer churn data,this article constructs a Bayesian and SVM model,and the possibility of customer churn prediction. In the process of experiment,this writer changes the size of the a-mount of data and characteristics of many fields,learns the results by means of experimental platform of clemen-tine 12 . 0 . Further comparative analysis of the five models comes to the conclusion:In the case that the attribute value is more,use the Bayesian-FS model;in the case that the attribute value is less,use the SVM polynomial kernel model,because the prediction accuracy and stability are better.
展开▼