Customer Lifetime Value Modeling via Two Stage Selected Trees Ensembles

The successful implementation of customer relationship management and product development need individual customer value assessment to generate sustained profit returns. As an important part of the customer relationship management, predicting future buyers and their monetary value allow companies to...

Full description

Saved in:
Bibliographic Details
Main Authors: Zardad Khan, Amjad Ali, Saeed Aldahmani, Henrik Nordmark, Berthold Lausen
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/11059931/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The successful implementation of customer relationship management and product development need individual customer value assessment to generate sustained profit returns. As an important part of the customer relationship management, predicting future buyers and their monetary value allow companies to tailor interactions, focus on high-value prospects, and increase customer profitability and retention. This paper proposes a system that adopts two sequential stages to calculate customer lifetime value based on their purchase behaviors. To handle the diversity and imbalance in customer data, we explore three decision tree selection strategies to form an overall efficient ensemble. Decision tree selection is done based on individual tree performance on independent samples, out-of-bag samples, and sub-samples for their inclusion in the final ensemble. The first stage predicts whether a customer will purchase in a future window, while the second stage estimates the monetary value of their future purchases. Model evaluation includes Brier score, accuracy, kappa, and sensitivity for the first stage, and unexplained variation, RMSE, Spearman correlation, and MAD for the second stage. Based on a thorough analysis of five customer transaction datasets that are further divided into power and sporadic customers data, the proposed ensemble methods surpass several other state-of-the-art methods, i.e., k nearest neighbor, weighted kNN, random forests, support vector machine and neural networks in majority of the cases considering both the stages.
ISSN:2169-3536