Deep Learning-Based Customer Lifetime Value Prediction in Imbalanced Data Scenarios: A Case Study

Analyzing and estimating customer lifetime value (LTV) is crucial for enterprises. Utilizing a game company's real needs and data as a case study, this paper proposes a two-stage model that models LTV in imbalanced data scenarios. The first stage model is designed for predicting user payments in imbalanced data scenarios, and the second stage model predicts the overall user LTV. The first stage model effectively addresses the issue of data imbalance, playing a crucial role in the subsequent models. The two-stage model can be combined to filter users with LTV greater than 0 at multiple granularities, like a funnel, and predict the overall user LTVto enable enterprises to optimize advertising expenditure, or each stage model can be used separately according to the actual situation, providing practical and flexible solutions to meet the diverse needs of enterprises in various scenarios. The experimental results of a case study show that the proposed method is a new approach for predicting LTV in imbalanced data scenarios, which is of great practical significance for enterprises.