Customer Churn Combination Prediction Model Based on Convolutional Neural Network and Gradient Boosting Decision Tree

Abstract

In order to improve the hit ratio of lost customers in telecom industry, a combination prediction model of customer churn based on one-dimensional convolutional neural network(1DCNN) and gradient boosting decision tree(GBDT) is proposed. Firstly, customer data is fed into 1DCNN model, which uses one-dimensional convolution to automatically extract customer features and then predicts customer churn through full connection layer. If the prediction result of 1DCNN model is churn, the result is directly output. If the prediction result is non-churn, the customer data will be re-introduced into GBDT model for second forecast, and the new prediction result will be output. Experiments on two publicly available telecom customer data set show that the proposed combined model significantly improves the recall rate and F1 score of customer churn prediction.