Title: An efficient covariate selection method for a Pareto-type model and machine learning in customer inactivity prediction

Authors: Shao-Ming Xie

Addresses: Nanlin Core Area, 3rd Floor, Building 7, Business Office Building, Jianyang District, Nanping City, Fujian Province, China

Abstract: Customer inactivity prediction is a key component in customer base maintenance, especially under cross-border e-commerce (CBEC) with a heterogeneous and complicated customer base on a global scale. This research harnesses genetic algorithm (GA) and an uplifting idea to set up a novel covariate selection method for customer inactivity prediction, empirically demonstrating that the proposed scheme is effective and robust. The model with selected covariates has indiscriminate or even better predictability over the model with all covariates or without any covariates, meaning that covariate selection is critical in modelling. The Pareto/NBD (Abe) model shows consistent predictability versus random forest on different samples, denoting the parametric model sustains more variations than the observation-based model. The proposed scheme selects these covariates that relate to the behavioural hypothesis of the Pareto/NBD (Abe) model. Lastly, random forest needs more diversified covariates to depict customer inactive status from different angles.

Keywords: customer inactivity prediction; genetic algorithm; GA; covariate selection; Pareto/NBD (Abe) model; random forest; cross-border e-commerce; CBEC.

DOI: 10.1504/IJTMKT.2026.150500

International Journal of Technology Marketing, 2026 Vol.20 No.1, pp.50 - 63

Accepted: 06 Jul 2025
Published online: 15 Dec 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article