Time series for early churn detection: Using similarity based classification for dynamic networks

作者：

Highlights：

• A novel way to featurize customer networks and capture their dynamic behavior.

• Adapted Similarity Forests for multivariate time series to predict customer churn.

• Similarity Forests applied to time series data outperform the 1-NN benchmark.

• The proposed method is better than traditional methods at detecting churn early.

摘要

•A novel way to featurize customer networks and capture their dynamic behavior.•Adapted Similarity Forests for multivariate time series to predict customer churn.•Similarity Forests applied to time series data outperform the 1-NN benchmark.•The proposed method is better than traditional methods at detecting churn early.

论文关键词：Multivariate time series,Churn prediction,Call detail records,Time series classification,Social networks,Dynamic networks

论文评审过程：Received 29 January 2018, Revised 13 March 2018, Accepted 3 April 2018, Available online 6 April 2018, Version of Record 13 April 2018.

论文官网地址：https://doi.org/10.1016/j.eswa.2018.04.003