A hybrid model for improving customer lifetime value prediction using stacking ensemble learning algorithm
A significant challenge that analysts and marketing managers often face is predicting future customer buying behavior. Identifying customers who are likely to make purchases down the line and estimating how much they will spend can help companies create more effective marketing campaigns and special...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-05-01
|
| Series: | Computers in Human Behavior Reports |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2451958825000314 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | A significant challenge that analysts and marketing managers often face is predicting future customer buying behavior. Identifying customers who are likely to make purchases down the line and estimating how much they will spend can help companies create more effective marketing campaigns and special offers that not only boost profits but also improve the overall customer experience and contribute to building lasting relationships. This paper outlines a comprehensive, detailed methodology for the forecasting and analysis of customer behavior using advanced predictive models and machine learning techniques. This framework can assist an organization in making more strategic and data-driven decisions to enhance both performance and profitability while securing improved customer satisfaction and long-term loyalty. The primary goal of this research is to increase the accuracy of Customer Lifetime Value (CLV) predictions and to provide deeper insights into customer behaviors. To achieve this objective, a combination of four key metrics is employed, namely: predicted purchases estimated using the BG-NBD model, predicted average value calculated through the Gamma-Gamma model, customer clustering labels assigned via the K-means model, and the customer status index derived from the Markov chain model. These essential metrics are subsequently integrated into the dataset. In order to further improve the accuracy of the predictions, this research uses a Stacking Ensemble model that combines four algorithms, i.e., Elastic Net, Random Forest, XGBoost, and SVM. The results demonstrate that integrating those features and using the Stacking Ensemble model substantially increases the prediction accuracy and decreases the errors. Sensitivity analysis and feature importance also prove the effectiveness of the method. |
|---|---|
| ISSN: | 2451-9588 |