Deep neural networks: predictive research on customer turnover caused by enterprise marketing problems

Abstract

Customer turnover prediction can assist enterprises in identifying potential lost customers early and formulating marketing strategies to retain them. This paper used telecom enterprise A as an illustrative example for customer turnover prediction. A balanced dataset was obtained through the synthetic minority oversampling technique (SMOTE) algorithm. Feature selection was conducted using the IV value. Additionally, the Inception v1 structure was optimized based on a deep neural network to design a deep convolutional neural network (CNN). Experiments were performed on the dataset of telecom enterprise A and the customer turnover datasets from Kaggle. On the Kaggle datasets, the deep CNN demonstrated superior classification performance compared to conventional approaches such as random forest (RF) and XGBoost. It exhibited a higher recall rate,  score, and area under the curve (AUC) value. The dataset of telecom enterprise A enhanced the prediction effectiveness of the deep CNN after processing by the SMOTE algorithm, and a recall rate of 0.97, a  score of 0.98, and an AUC value of 0.98 were achieved. These results show the reliability of the deep CNN for customer turnover prediction and its practical applicability.