基于时序模式匹配的k-近邻分类在流失预测中的应用

Application of k-nearest Neighbors Classification Based on Time-series Pattern Matching in Churn Prediction

为了解决电信行业中如何预测用户流失的问题,该文提出了一种基于时序模式匹配的k-近邻分类方法。与传统的预测方法(如基于决策树的方法)相比,该方法分类时序数据时,不需要将时序数据离散化为非时序数据。该文详细描述了算法的设计以及在真实的电信数据上的应用。与C4.5方法的实验结果比较,表明了该方法有效地保留了时序的完整性,在一定程度上提高了预测准确率。

To solve the problem of churn prediction in telecom, this paper proposes a k nearest neighbors classification method based on time-series pattern matching. This method integrates time-series pattern matching and k-nearest neighbor classification to predict the class of unclassified data. Compared with the conventional prediction methods （such as decision-tree based method）, the method can make prediction without discrete transition for time-series. This paper describes in detail the design and the application to the real telecommunication data of the algorithm. The comparison of experimental results between the method of this paper and C4.5 indicates that the proposed method preserves the integrity of time series and can improve the accuracy of the prediction.