Machine learning aplicado al análisis de riesgo de crédito
Loading...
Identifiers
Publication date
Reading date
Authors
Collaborators
Advisors
Tutors
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Share
Department/Institute
Abstract
Este trabajo tiene como objetivo analizar la aplicabilidad de técnicas de Machine Learning en el análisis de riesgo crediticio, con un enfoque particular en la selección de variables relevantes y la optimización de modelos predictivos. La revisión de la literatura destaca que estas técnicas superan a los modelos tradicionales en la predicción de riegos y la reducción de pérdidas. Además, se identifica el reciente interés por la aplicabilidad de enfoques no convencionales basados en métodos Deep Learning que, junto con datos alternativos no financieros, promueven la inclusión financiera. Como metodología se proponen dos algoritmos de optimización: Búsqueda Local (BL) y Cross Entropy (CE), los cuales se combinan con distintas técnicas de Machine Learning, que tienen como propósito potenciar el proceso que identifica combinaciones óptimas de variables, identificando las más influyentes en la predicción del riesgo crediticio. Los resultados muestran que el preprocesamiento de datos, incluyendo la imputación de valores y la eliminación de outliers es fundamental para aumentar la precisión de los modelos. El análisis comparativo entre BL y CE sugiere que ambos métodos mejoran el rendimiento de los modelos, aunque CE tiende a ofrecer una mayor flexibilidad en la selección de otras variables, obteniendo mejoras adicionales en el proceso de optimización. Además, se encuentra que los retrasos más prolongados en el pago de préstamos se constituye como los predictores claves del análisis de riesgo.
This thesis aims to analyze the applicability of Machine Learning tecniques in credit risk analysis, with a particular focus on the selection of relevant variables and optimization of predictive models. The literature review highlights that these techniques outperform traditinal methods in risk prediction and loss reduction. Additionally, there is growing interest in the applicability of unconventional approaches based on Deep Learning methods, which, togheter with non-financial alternative data, promote financial inclusion. As methodology, two optimization algorithms are proposed: Local Search (BL) and Cross Entropy, combined with various Machine Learning techniques to enhance the process to indentifying optimal combinations of variables, focusing on the most influential ones in credit risk prediction. The results show that data preprocessing, including value imputation and outlier removal, crucial to improving model accuracy. The comparative analisys between BL and CE suggests that both methods improve model performance, although CE offers greater flexibility in selecting additional variables, leading to further optimization improvements. Moreover, it was found that longer loan payment delays are key predictors in credit risk analisys.
This thesis aims to analyze the applicability of Machine Learning tecniques in credit risk analysis, with a particular focus on the selection of relevant variables and optimization of predictive models. The literature review highlights that these techniques outperform traditinal methods in risk prediction and loss reduction. Additionally, there is growing interest in the applicability of unconventional approaches based on Deep Learning methods, which, togheter with non-financial alternative data, promote financial inclusion. As methodology, two optimization algorithms are proposed: Local Search (BL) and Cross Entropy, combined with various Machine Learning techniques to enhance the process to indentifying optimal combinations of variables, focusing on the most influential ones in credit risk prediction. The results show that data preprocessing, including value imputation and outlier removal, crucial to improving model accuracy. The comparative analisys between BL and CE suggests that both methods improve model performance, although CE offers greater flexibility in selecting additional variables, leading to further optimization improvements. Moreover, it was found that longer loan payment delays are key predictors in credit risk analisys.
Description
Bibliographic citation
Collections
Endorsement
Review
Supplemented By
Referenced by
Creative Commons license
Except where otherwised noted, this item's license is described as Attribution-NonCommercial-NoDerivatives 4.0 International










