A Feature Extraction Method Based on Stacked Auto-Encoder for Telecom Churn Prediction.

Ruiqi Li,Peng Wang,Zonghai Chen
DOI: https://doi.org/10.1007/978-981-10-2663-8_58
2016-01-01
Abstract:Customer churn prediction is a key problem to customer relationship management systems of telecom operators. Efficient feature extraction method is crucial to telecom customer churn prediction. In this paper, stacked auto-encoder is introduced as a nonlinear feature extraction method, and a new hybrid feature extraction framework is proposed based on stacked auto-encoder and Fisher's ratio analysis. The proposed method is evaluated on datasets provided by Orange, and experimental results verify that it is authentically able to enhance the performance of prediction models both on AUC and computing efficiency.
What problem does this paper attempt to address?