New Approach to Telecom Churn Prediction Based on Transformers

Abstract

Retaining customers costs more than acquiring new ones to telecommunications providers. Predicting churn has therefore become an extremely critical issue for leading service providers worldwide. As a result, major investments are increasingly being made in the design of innovative strategies to combat churn. Machine learning is part of the efforts performed nowadays in this field. In the present paper, we have built a new approach to predict with a high accuracy the churn, our model is based on transforming features into Radar Chart image and then apply a transformer architecture for classification. The model we suggested scored a good result in terms of accuracy, 81%.