Bibliographic citations
This is an automatically generated citacion. Modify it if you see fit
Perales, H., (2024). Modelo de calificación basado en redes neuronales para la predicción de riesgo crediticio: caja financiera [Universidad de Lima]. https://hdl.handle.net/20.500.12724/20766
Perales, H., Modelo de calificación basado en redes neuronales para la predicción de riesgo crediticio: caja financiera []. PE: Universidad de Lima; 2024. https://hdl.handle.net/20.500.12724/20766
@misc{renati/234852,
title = "Modelo de calificación basado en redes neuronales para la predicción de riesgo crediticio: caja financiera",
author = "Perales Paz, Hercy Antonio",
publisher = "Universidad de Lima",
year = "2024"
}
Title: Modelo de calificación basado en redes neuronales para la predicción de riesgo crediticio: caja financiera
Authors(s): Perales Paz, Hercy Antonio
Advisor(s): Guzman Jimenez, Rosario Marybel
OCDE field: https://purl.org/pe-repo/ocde/ford#2.02.04
Issue Date: 2024
Institution: Universidad de Lima
Abstract: En el ámbito de las microfinanzas, surge la necesidad imperante de anticipar el riesgo crediticio de los clientes debido
al crecimiento de los incumplimientos de pagos y la morosidad. Este estudio se enfoca en la predicción del riesgo
crediticio tanto para individuos como para negocios, mediante un modelo de calificación basado en análisis de
ecuaciones estructurales y redes neuronales. El objetivo es discernir entre clientes de buen y mal perfil crediticio. Se
empleó una base de datos construida a partir de los registros de desembolsos de préstamos efectuados por una entidad
financiera de provincia durante el período 2022-2023. Para el análisis mediante ecuaciones estructurales, se seleccionó
una muestra de 382 analistas de crédito y se evaluaron 28 variables distribuidas en 5 dimensiones (características del
cliente, nivel de endeudamiento, datos demográficos, aspectos operativos y predicción del riesgo del cliente). Se
obtuvo un coeficiente de determinación (R2) del 6% para las variables propuestas en cada dimensión. Por otro lado,
se implementó una red neuronal con 28 neuronas de entrada, tres capas ocultas y una neurona de salida. Los pesos se
ajustaron adaptativamente en función de la magnitud de la derivada del error durante el proceso de aprendizaje. Este
enfoque arrojó una precisión del 91.5%, superando los resultados previamente reportados en la literatura para este
contexto específico.
In the context of microfinance, there is a need to predict the credit risk of a client given the increase in payment defaults and arrears. The objective of this study is to predict the credit risk of individuals and businesses in a credit assessment process through a scoring model based on structural equation analysis and neural networks that can distinguish between a good and a bad client. A database was constructed from loan disbursement records of a provincial financial institution during the period 2022-2023. For the application of the structural equation technique, a sample of 382 credit analysts was taken to analyse 28 variables distributed in 5 dimensions (client characteristics, client indebtedness, client demographic characteristics, operational characteristics and client risk prediction). An R2 of 6% was obtained for the variables proposed for each dimension. A neural network with 28 input neurons, three hidden layers and one output neuron was used and the weights were adaptively adjusted based on the magnitude of the error derivative during the learning process. An accuracy of 91.5% was achieved, which is superior to the literature review results for this context.
In the context of microfinance, there is a need to predict the credit risk of a client given the increase in payment defaults and arrears. The objective of this study is to predict the credit risk of individuals and businesses in a credit assessment process through a scoring model based on structural equation analysis and neural networks that can distinguish between a good and a bad client. A database was constructed from loan disbursement records of a provincial financial institution during the period 2022-2023. For the application of the structural equation technique, a sample of 382 credit analysts was taken to analyse 28 variables distributed in 5 dimensions (client characteristics, client indebtedness, client demographic characteristics, operational characteristics and client risk prediction). An R2 of 6% was obtained for the variables proposed for each dimension. A neural network with 28 input neurons, three hidden layers and one output neuron was used and the weights were adaptively adjusted based on the magnitude of the error derivative during the learning process. An accuracy of 91.5% was achieved, which is superior to the literature review results for this context.
Link to repository: https://hdl.handle.net/20.500.12724/20766
Discipline: Ingeniería de Sistemas
Grade or title grantor: Universidad de Lima. Facultad de Ingeniería
Grade or title: Ingeniero de Sistemas
Juror: Pendiente
Register date: 21-Jun-2024
This item is licensed under a Creative Commons License