Predicción de la emisión de tarjetas de crédito por instituciones financieras del Ecuador con técnicas de aprendizaje automático

Abstract

CONDITION FOR PUBLICATION OF PROJECT. The issuance of credit cards plays a crucial role for financial institutions, as it allows them to generate revenue through fees, interest, and installments, while also providing benefits to both customers and the institutions. For these entities, accurately forecasting credit card demand is essential, as it enables them to optimize resource allocation, design more effective marketing campaigns, and manage financial risks more efficiently. A predictive pipeline based on time series analysis was developed using an AutoML approach, which automates the training, evaluation, and selection of the most appropriate model using metrics such as the coefficient of determination (R²) and mean squared error (MSE). The results showed R² values of 0.5294 for the aggregate of all entities, 0.7822 for Diners Club, and 0.6127 for Produbanco, with an overall average of 64.8%, which validates the model’s ability to capture the behavior and trends of each institution. Based on these results, an interactive web tool was developed, allowing financial institutions to predict future credit card demand, thus optimizing strategic decision-making. The tool includes an interactive dashboard through which users can explore various data features and visualize the performance of the entities. Additionally, it facilitates the comparison of these results with those of competitors, improving the effectiveness of commercial and marketing policies by providing a clearer view of market trends. Keywords: Credit Card Issuance, AutoML, Time Series, Prediction