基于特征选择和SVM的电信客户离网预测

Prediction on customer leaving the telecom network based on feature selection and SVM

针对数据挖掘算法在预测电信客户离网时存在的过拟合问题,提出一种基于特征选择和支持向量机的电信客户离网预测算法。将原始的电信数据分别进行数据缺失值填充、数据冗余识别、数据结构化和数据归一化等预处理,得到利于分析处理的规范性数据;利用信息增益完成特征选择,提取影响客户离网的主要因素,降低数据维度,防止出现过拟合现象。将经过特征选择后的数据作为支持向量机算法的输入数据对客户是否离网进行分类,预测客户是否存在离网行为。测试结果表明,该算法预测离网客户的正确率为86%,提升了离网客户预测准确率。

To solve the overfitting problem on predicting the customer s leaving the telecom network for data mining algorithms, a new algorithm based on feature selection and support vector machine is proposed in this paper. Original telecommunication data are processed through data loss, data redundancy identification and data structure to obtain the normalized data. Using information gain for feature selection, the main factors affecting customer out of network are extracted to remove irrelevant or redundant features and then to reduce the data dimension and prevent overfitting. The data after feature selection is then used as the input data of the SVM algorithm to classify whether the customer is out of network, to predict whether the customer has behaviours of potentially leaving the telecom network. Prediction results using this algorithm show that the accuracy rate of leaving the telecom network is 86%, and thus show this algorithm can improve prediction accuracy on the customer leaving the telecom network.