Core Concepts
The authors propose a novel framework for predicting customer churn in the telecommunications industry while preserving data privacy using Generative Adversarial Networks (GANs) with differential privacy and adaptive Weight-of-Evidence (aWOE) transformation.
Stats
The GANs-aWOE based Naïve Bayes model achieved an F-measure of 87.1%.
The proposed approach demonstrated a prediction enhancement of up to 28.9% and 27.9% in terms of accuracy and F-measure, respectively, compared to previous studies.
The study used three publicly available datasets with sample sizes of 100,000, 7,043, and 5,000.
The privacy budget parameter (ϵ) for differential privacy was set to 10.
Quotes
"Protecting the privacy of data is difficult when data owners outsource the machine learning task to a cloud service provider."
"To the best of our knowledge, GANs based privacy preserving customer churn prediction has not yet been studied in the literature."
"Our main objective is to preserve data privacy while performing third-party computation without sacrificing performance."