Comparación de modelos de aprendizaje de máquina en la predicción del incumplimiento de pago en el sector de las microfinanzas
No hay miniatura disponible
Fecha
2021-06-24
Autores
Título de la revista
ISSN de la revista
Título del volumen
Editor
Pontificia Universidad Católica del Perú
DOI
Resumen
Las instituciones financieras dedicadas a las Microfinanzas brindan sus servicios a un público
objetivo que en su mayoría presentan bajos recursos económicos y/o cuyo acceso a los
sistemas bancarios tradicionales es limitado, estas instituciones al desarrollarse en un
contexto poco favorable los riesgos de incumplimiento en los pagos son mayores en
comparación a la banca tradicional. Por tanto, se exige hacer una evaluación económica
financiera con mayor grado de detalle, requiriendo para tal fin la participación de un
experto del negocio que basado en información obtenida y pericia propia determine si el
potencial cliente será un buen pagador. Esta forma de evaluar a un cliente ha evolucionado
en el sector financiero en los últimos años, esto debido en gran medida a la aplicación de
tecnologías como la inteligencia artificial y el aprendizaje de máquina, ofreciendo una
singularidad que es la capacidad de aprender de los datos, demandando menos esfuerzo y
participación humana, y redituando mayores niveles de precisión. Se presentan en este
artículo los resultados de la experimentación realizada con los siguientes modelos de
aprendizaje de maquina: Regresión Logística, XGBoost, Random Forest, Gradient Boosting,
Perceptron Multicapa (MLP) y algoritmos de aprendizaje profundo para la predicción del
incumplimiento de pagos, aplicándose técnicas de balanceo de submuestreo y
sobremuestreo, incluida la técnica de SMOTE. Así mismo, se aplicó la técnica de One Hot
Encoding para el tratamiento de variables categóricas. Los diferentes modelos de
aprendizaje de maquina se aplicaron a un conjunto de datos proporcionado por una
institución peruana líder en el sector de las microfinanzas, reportando los mejores
resultados el modelo XGBoost, con una exactitud de 97.53% y un F1-Score de 0.1278.
The financial institutions dedicated to Microfinance offer their services to a target audience that, for the most part, has low economic resources and/or whose access to traditional banking systems is limited, these institutions to develop in an unfavorable context the risks of non-compliance in the payments are greater compared to traditional banking, therefore it is required to make a financial economic evaluation with a greater degree of detail, requiring for this purpose the participation of a business expert that based on information obtained and own expertise determine if the potential client will be a good payer, this way of evaluating a customer has evolved in the financial sector in recent years, this largely due to the application of technologies such as artificial intelligence and machine learning, offering a uniqueness that is the ability to learn from the data, demanding less effort and human participation mana, and yielding higher levels of accuracy. This article presents the results of the experimentation carried out with the following machine learning models: Logistic Regression, XGBoost, Random Forest, Gradient Boosting, Multilayer Perceptron (MLP) and deep learning algorithms for the prediction of non-payment, applying subsampling and oversampling balancing techniques, including the SMOTE technique, and the One Hot Encoding technique was applied for the treatment of categorical variables. The different models of machine learning were applied to a data set provided by a leading Peruvian institution in the microfinance sector, with the XGBoost model reporting the best results, with an accuracy of 97.53% and an F1-Score of 0.1278.
The financial institutions dedicated to Microfinance offer their services to a target audience that, for the most part, has low economic resources and/or whose access to traditional banking systems is limited, these institutions to develop in an unfavorable context the risks of non-compliance in the payments are greater compared to traditional banking, therefore it is required to make a financial economic evaluation with a greater degree of detail, requiring for this purpose the participation of a business expert that based on information obtained and own expertise determine if the potential client will be a good payer, this way of evaluating a customer has evolved in the financial sector in recent years, this largely due to the application of technologies such as artificial intelligence and machine learning, offering a uniqueness that is the ability to learn from the data, demanding less effort and human participation mana, and yielding higher levels of accuracy. This article presents the results of the experimentation carried out with the following machine learning models: Logistic Regression, XGBoost, Random Forest, Gradient Boosting, Multilayer Perceptron (MLP) and deep learning algorithms for the prediction of non-payment, applying subsampling and oversampling balancing techniques, including the SMOTE technique, and the One Hot Encoding technique was applied for the treatment of categorical variables. The different models of machine learning were applied to a data set provided by a leading Peruvian institution in the microfinance sector, with the XGBoost model reporting the best results, with an accuracy of 97.53% and an F1-Score of 0.1278.
Descripción
Palabras clave
Aprendizaje automático (Inteligencia artificial), Microfinanzas
Citación
item.page.endorsement
item.page.review
item.page.supplemented
item.page.referenced
Licencia Creative Commons
Excepto se indique lo contrario, la licencia de este artículo se describe como info:eu-repo/semantics/openAccess