Automating Credit Card Limit Adjustments Using Machine Learning
Diego Pestana, Enrique Areyan Viqueira
TL;DR
The paper tackles automating credit card limit adjustments (CLAD) in Venezuela by adopting cost-sensitive learning to replace manual committee decisions, addressing misclassification costs and an increasing cardholder base. It compares neural networks and XGBoost under a grid-search, 10-fold cross-validation framework and uses Cohen's kappa to measure agreement with committee outcomes, ultimately selecting XGBoost for its balance of accuracy, cost, and interpretability. The XGBoost model achieves near-parity with the best neural network in accuracy while reducing total cost and providing greater explainability, yielding a practical, scalable solution for CLAD that has been adopted by the national bank's risk committee. Despite the small dataset, the approach demonstrates a viable path to automate CLAD with transparency and potential applicability to other banks, with future work focused on scaling data and exploring online deployment.
Abstract
Venezuelan banks have historically made credit card limit adjustment decisions manually through committees. However, since the number of credit card holders in Venezuela is expected to increase in the upcoming months due to economic improvements, manual decisions are starting to become unfeasible. In this project, a machine learning model that uses cost-sensitive learning is proposed to automate the task of handing out credit card limit increases. To accomplish this, several neural network and XGBoost models are trained and compared, leveraging Venezolano de Credito's data and using grid search with 10-fold cross-validation. The proposed model is ultimately chosen due to its superior balance of accuracy, cost-effectiveness, and interpretability. The model's performance is evaluated against the committee's decisions using Cohen's kappa coefficient, showing an almost perfect agreement.