@Article{cmes.2023.029023,
AUTHOR = {Yaling Xu, Congjun Rao, Xinping Xiao, Fuyan Hu},
TITLE = {Novel Early-Warning Model for Customer Churn of Credit Card Based on GSAIBAS-CatBoost},
JOURNAL = {Computer Modeling in Engineering \& Sciences},
VOLUME = {137},
YEAR = {2023},
NUMBER = {3},
PAGES = {2715--2742},
URL = {http://www.techscience.com/CMES/v137n3/53740},
ISSN = {1526-1506},
ABSTRACT = {As the banking industry gradually steps into the digital era of Bank 4.0, business competition is becoming
increasingly fierce, and banks are also facing the problem of massive customer churn. To better maintain their
customer resources, it is crucial for banks to accurately predict customers with a tendency to churn. Aiming at the
typical binary classification problem like customer churn, this paper establishes an early-warning model for credit
card customer churn. That is a dual search algorithm named GSAIBAS by incorporating Golden Sine Algorithm
(GSA) and an Improved Beetle Antennae Search (IBAS) is proposed to optimize the parameters of the CatBoost
algorithm, which forms the GSAIBAS-CatBoost model. Especially, considering that the BAS algorithm has simple
parameters and is easy to fall into local optimum, the Sigmoid nonlinear convergence factor and the lane ight
equation are introduced to adjust the fixed step size of beetle. Then this improved BAS algorithm with variable
step size is fused with the GSA to form a GSAIBAS algorithm which can achieve dual optimization. Moreover, an
empirical analysis is made according to the data set of credit card customers from Analyttica ocial platform. The
empirical results show that the values of Area Under Curve (AUC) and recall of the proposed model in this paper
reach 96.15% and 95.56%, respectively, which are significantly better than the other 9 common machine learning
models. Compared with several existing optimization algorithms, GSAIBAS algorithm has higher precision in the
parameter optimization for CatBoost. Combined with two other customer churn data sets on Kaggle data platform,
it is further verified that the model proposed in this paper is also valid and feasible.},
DOI = {10.32604/cmes.2023.029023}
}