多粒度时序特征在离网预测中的应用

电信运营商为了发现可能离网的客户，针对不同的场景研究开发了多种离网预测模型。目前的离网预测模型首先选择一种时间粒度抽取特征，之后使用机器学习算法对抽取的数据建模。这类方法只考虑了模型对分类性能的影响，没有充分考虑数据的作用。针对上述问题，提出一种使用多种时间粒度抽取特征的方法，并尝试在模型训练的不同阶段对不同粒度的特征进行融合。实验结果表明，使用多种粒度抽取特征训练出来的模型性能会明显优于使用单一粒度抽取特征的模型。

Application of multi-grain temporal features in churn prediction

Telecom operators have developed multiple churn prediction models to find potential users for different scenes. The present churn prediction models firstly select a kind of time granularity to extract features, then model the extracted data using machine learning algorithm. Such approaches only consider the influence of the model on classification performance, but the role of data is not fully considered. To solve this problem, this paper proposed a method which extracts multi-grain temporal features, and try to integrate different granularity features at different training phases. Experimental results show that the performance of the model trained with multi-grain features is obviously superior than that trained with single granularity features.