Desarrollo de un sistema predictivo para la asignación óptima de cupos de crédito en una empresa de nutrición animal

Abstract

CONDITION FOR PUBLICATION OF PROJECT. This project focuses on the development of a predictive model that, at invoice issuing, estimates the risk of non‑payment in a company’s accounts receivable. The hypothesis posits that by applying machine learning techniques it is possible to identify customers at elevated risk of default before it occurs, shifting collections management from a reactive to a proactive approach. The motivation stems from the need to reduce the costs and effort involved in recovering overdue accounts, while optimizing cash flow and improving the organization’s financial health. Simulated historical data inspired by the animal nutrition sector was used, with structures and variables like those typically managed by companies in the industry. Data preprocessing included record cleaning, generation of additional features, removal of redundant variables through correlation analysis, and class balancing via SMOTE, ensuring a robust and representative dataset for model training. Logistic regression, decision trees, random forest, and LightGBM algorithms were trained using time‑based validation and grid search for hyperparameter tuning. Performance was evaluated with metrics such as precision, recall, and F1‑score, with particular emphasis on the ability to detect true defaults. The random forest model demonstrated a superior balance between accuracy and generalization, emerging as the most robust choice for production deployment. Finally, an interactive Gradio prototype was developed that, upon selecting a customer and optionally adjusting their payment parameters, calculates in real time the probability of default for a new invoice. This tool is proposed as a strategic asset for proactive receivables management, reducing delinquency risk and strengthening the company’s financial sustainability. Keywords: predictive model, default risk, machine learning, receivables management, financial risk.