Customer churn prediction for a webcast platform via a voting-based ensemble learning model with Nelder-Mead optimizer

Abstract

This article studies the application for customer churn prediction on webcast. Predicting churn customers become an urgent need in webcast industry because the market is getting saturated and identifying potential churn customers and developing recall marketing strategies can save companies significant costs. Despite the importance of customer churn prediction in many fields, little prior academic attention has been attached to the webcast area. To address this gap, We apply an ensemble learning method to build a binary classification model for customer churn prediction. Our proposed model uses a weighted voting ensemble method and the Nelder-Mead optimal algorithm with a specific focus on the speed of Internet customers’ mobility, extracting high-dimensional features from time series data to incorporate more detailed customer behavior information. In addition, a new customer churn indicator based on time decline is introduced to more accurately define churned customers in the training data. The experimental data is collected from a webcast application developed by a Chinese Internet company. Experimental evaluations show that compared to the traditional ensemble models, our proposed model is operationally efficient and outperforms other approaches, providing valuable insights for companies to intervene with churned customers and adopt targeting retention interventions.