Uso de modelos no supervisados y modelos supervisados para segmentar clientes y predecir su conducta de pago en Billingsof Punto Pymes.

Abstract

Abstract: The management of overdue accounts represents a critical challenge for service-oriented companies, particularly when advanced analytical tools are not used to anticipate default risk. This study aims to implement a mixed analytical model that combines unsupervised and supervised machine learning techniques to segment customers and predict their payment behavior at BillingSof Punto Pymes. Using historical data related to billing, payment behavior, and customer characteristics, exploratory data analysis was conducted, followed by dimensionality reduction through Principal Component Analysis (PCA) and customer segmentation using K-Means clustering to identify differentiated risk profiles. Subsequently, supervised models, including logistic regression and decision trees, were trained to estimate the probability of default, with performance evaluated through metrics such as AUC, precision, and recall. The results demonstrate that the proposed approach improves risk classification, optimizes the allocation of collection strategies, and supports more efficient, data-driven decision-making, contributing to improved financial sustainability and customer relationship management.