Scholarlink

Accurately predicting customer lifetime value (LTV) is essential for optimizing marketing strategies and enhancing customer relationship management. This paper presents a machine learning project implemented within a corporate setting to forecast customer LTV. The project explored a variety of machine learning algorithms including Linear Regression, Decision Trees, Extreme Gradient Boosting (XGBoost), and Artificial Neural Networks (ANNs), alongside multiple encoding techniques for handling categorical features. A rigorous evaluation was conducted using fivefold time-series cross-validation to ensure the robustness and reliability of the predictions in a temporal context. The performance of the algorithms measured by Normalized Root Mean Square Error (NRMSE) had values ranging from 1.740 to 1.776. After a comprehensive analysis, XGBoost was selected for deployment based on its performance and operational efficiency. The deployed model is currently active, contributing to essential marketing decisions. This paper outlines the project’s approach, findings, and practical insights and discusses the implementation challenges and achievements. Through this study, we aim to contribute to the ongoing discussion on the practical application of machine learning in business environments, offering a case study that bridges theoretical constructs with real-world utility.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing Customer Lifetime Value Prediction in a Service Marketplace Through Machine Learning: An Empirical Comparison of Algorithms

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Seyf Kazamel,
  • Omar Abdelaty,
  • Ali Alıcı

摘要

Accurately predicting customer lifetime value (LTV) is essential for optimizing marketing strategies and enhancing customer relationship management. This paper presents a machine learning project implemented within a corporate setting to forecast customer LTV. The project explored a variety of machine learning algorithms including Linear Regression, Decision Trees, Extreme Gradient Boosting (XGBoost), and Artificial Neural Networks (ANNs), alongside multiple encoding techniques for handling categorical features. A rigorous evaluation was conducted using fivefold time-series cross-validation to ensure the robustness and reliability of the predictions in a temporal context. The performance of the algorithms measured by Normalized Root Mean Square Error (NRMSE) had values ranging from 1.740 to 1.776. After a comprehensive analysis, XGBoost was selected for deployment based on its performance and operational efficiency. The deployed model is currently active, contributing to essential marketing decisions. This paper outlines the project’s approach, findings, and practical insights and discusses the implementation challenges and achievements. Through this study, we aim to contribute to the ongoing discussion on the practical application of machine learning in business environments, offering a case study that bridges theoretical constructs with real-world utility.