摘要
本文提出了一种新的基于多分类器动态选择与优化集成的电信客户流失预测集成模型.首先使用K均值聚类算法对训练集样本进行分区;然后分别使用Naive-Bayes算法、多层感知机算法和J48算法构建各分区客户流失预测子分类器;最后对各分区子分类器进行线性集成,并使用人工蜂群算法优化其集成权重.当测试样本由聚类算法判断出其归属区域后,再分别使用分区子分类器进行预测,最后使用优化权重进行线性集成.实验结果表明:动态集成模型优于单模型;基于人工蜂群算法优化集成模型优于其它集成模型.
In this paper,a new customer churn prediction model used in telecommunication is put forward,which is based on the dynamic selection and optimizing integration of multiple classifiers.Firstly,the training set samples are clustered into multiple subareas in character space using K-means clustering algorithm.Then,customer churn prediction sub-classifiers are established based on the samples in the subareas using Naive-Bayes algorithm(NB),multilayer perceptron(MP) and J48 algorithm(J48),respectively.Finally,the sub-classifiers are integrated linearly,and the weighted coefficients are optimized using artificial bee colony algorithm(ABCA).After the test samples are clustered into corresponding subareas using the trained K-means clustering algorithm,the prediction is processed using the corresponding sub-classifier.The ultimate predicted results of test samples are obtained by linearly integrating the predicted results with the optimized weights.The experiment results show that the dynamic integration of models excels the all single models,and the predict precision based on the model integrated by ABCA excels these based on the models integrated by other algorithms.
出处
《系统工程学报》
CSCD
北大核心
2010年第5期703-711,共9页
Journal of Systems Engineering
基金
国家自然科学基金资助项目(70801021)
中国博士后科学基金资助项目(20080431276)
教育部人文社会科学资助项目(08JC630019)
关键词
客户流失预测
多分类器动态选择
多分类器优化集成
人工蜂群算法
customer churning prediction
dynamic selection of multiple classifiers
optimizing integration of multiple classifiers
artificial bee colony algorithm(ABCA).