Modelo de Clasificación de clientes morosos de una empresa de telefonía empleando Xgboost y Ligthgbm, Lima 2023

• Autores: Gargate Obregón, Samuel Gustavo, Ponce Aruneri, Mar´ıa Estela
• Formato: artículo
• Fecha de Publicación: 2025
• Institución: Universidad Nacional Mayor de San Marcos
• Repositorio: Revistas - Universidad Nacional Mayor de San Marcos
• Lenguaje: español
• OAI Identifier: oai:revistasinvestigacion.unmsm.edu.pe:article/32268
• Enlace del recurso: https://revistasinvestigacion.unmsm.edu.pe/index.php/matema/article/view/32268
• Nivel de acceso: acceso abierto
• Materia: Clasificación; XGBoost; LightGBM; Clientes morosos; Empresa de telefonía celular; Multiple correspondence analysis; municipal district management level; conservation of green areas in public spaces

The objective was to compare the classification of delinquent customers of a telephone company in Lima in 2022 using XGBoost and LightGBM. The research adopts a quantitative approach using statistical tools to analyze factors influencing postpaid customer delinquency. The research is explanatory in scope, employing statistical techniques to identify influential factors, as well as optimizing the classification through machine learning algorithms and neural networks. The design is non-experimental and cross-sectional, using a telephone company’s customer database. The study population includes postpaid customers meeting certain inclusion criteria, identifying 22,510 customers who meet the requirements. Among the results, XGBoost was found to have an accuracy of 0.68, slightly better than LightGBM (0.67). XGBoost excels in delinquency indicators compared to LightGBM, where the F1-Score indicator offers a more balanced measure of performance: XGBoost scores 0.2592 versus LightGBM’s 0.1443.