论文部分内容阅读
根据已有的流失预测方法,提出新的流失预测方法解决数据挖掘中的非对称错分代价问题.该方法以传统C4.5决策树算法为基准分类器,融合代价调整方法实现代价敏感学习.相比之下,C4.5决策树算法仅是基于样本错分代价相同假定,建立了一种错分率最低而非总错分代价最低的预测模型.基于某电信企业的客户数据,及流失客户和非流失客户代价非对称的实际,实证研究结果表明,CS-C4.5通过调整流失类和非流失类样本的比例,大大降低了传统分类算法的样本错分总代价.该方法对于提高电信企业的核心竞争力具有重要的现实意义.
According to the existing loss prediction method, a new loss prediction method is proposed to solve the asymmetric misclassification cost in data mining, which uses the traditional C4.5 decision tree algorithm as a reference classifier and cost-adjusted learning method to realize cost-sensitive learning. In contrast, the C4.5 decision tree algorithm is based on the same assumption that the sample misclassification cost is the same, and a prediction model with the lowest misclassification rate and the lowest total mis-division cost is established.Based on the customer data of a telecom enterprise and the loss Customer and non-loss customers are asymmetric, the empirical research results show that CS-C4.5 greatly reduces the total misclassification cost of the traditional classification algorithm by adjusting the proportion of missing and non-losing samples, The core competitiveness of telecom enterprises has important practical significance.