Loading...
Publication
Inteligência artificial aplicada à avaliação de crédito bancário
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 1.77 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
A avaliação de crédito é uma ferramenta financeira, importante, para bancos e instituições finan ceiras determinarem se devem emitir o empréstimo para potenciais mutuários. A utilização de
inteligência artificial levou a um melhor desempenho dos modelos de avaliação de crédito. Várias
técnicas de machine learning baseadas em estatísticas foram empregues para esta tarefa, sendo
que a regressão logística é o padrão da indústria na modelação de risco de crédito. Porém, estu dos demonstram que algoritmos de ensemble, que principalmente podem ser divididos em bagging
ensembles e boosting ensembles, têm se mostrado muito promissores.
Esta tese tem como objetivo comparar diversos tipos de modelos de machine learning de forma
a determinar quais oferecem o melhor desempenho para a classificação de crédito bancário. Para
tal, este estudo irá realizar comparações com diversos tipos de modelos de classificação, desde
os modelos tradicionais, como Logistic regression (LR), Linear discriminant analysis (LDA) e
Artificial neural network (ANN), a modelos mais recentes como ensemble homogéneos, tais como
AdaBoost, Gradient-Boosted Decision Trees (GBDT), eXtreme Gradient Boosting (XGBoost),
Light Gradient Boosting Machine (LightGBM) e CatBoost, até modelos mais experimentais como
o caso de modelos ensemble heterogéneos.
A contribuição final desta tese será fornecer informação de que modelos de machine learning
atualmente mais se adequam a avaliação de crédito bancário, com intuito de substituir os métodos
tradicionais.
Credit assessment is an important financial tool for banks and financial institutions to determine whether to issue the loan to potential borrowers. The use of artificial intelligence led to a better performance of credit assessment models. Several statistical-based machine learning techniques were employed for this task, with logistic regression being the industry standard in credit risk modeling. However, studies show that ensemble algorithms, which mainly can be divided into bagging ensembles and boosting ensembles, have shown to be very promising. This thesis aims to compare different types of machine learning models in order to determine which ones offer the best performance for bank credit rating. To this end, this study will carry out comparisons with different types of classification models, from traditional models like Logistic regression (LR), Linear discriminant analysis (LDA) and Artificial neural network (ANN) to more recent models such as homogeneous ensemble like AdaBoost, Gradient-Boosted Decision Trees (GBDT), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM) and CatBoost to more experimental models such as the case of heterogeneous ensemble models. The final contribution of this thesis will be to provide information on which machine learning models are currently best suited to bank credit assessment, in order to replace traditional methods.
Credit assessment is an important financial tool for banks and financial institutions to determine whether to issue the loan to potential borrowers. The use of artificial intelligence led to a better performance of credit assessment models. Several statistical-based machine learning techniques were employed for this task, with logistic regression being the industry standard in credit risk modeling. However, studies show that ensemble algorithms, which mainly can be divided into bagging ensembles and boosting ensembles, have shown to be very promising. This thesis aims to compare different types of machine learning models in order to determine which ones offer the best performance for bank credit rating. To this end, this study will carry out comparisons with different types of classification models, from traditional models like Logistic regression (LR), Linear discriminant analysis (LDA) and Artificial neural network (ANN) to more recent models such as homogeneous ensemble like AdaBoost, Gradient-Boosted Decision Trees (GBDT), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM) and CatBoost to more experimental models such as the case of heterogeneous ensemble models. The final contribution of this thesis will be to provide information on which machine learning models are currently best suited to bank credit assessment, in order to replace traditional methods.
Description
Keywords
Artificial intelligence Machine learning Homogeneous ensembles Heterogeneous ensembles Supervised learning Credit scoring Crédito bancário Inteligência artificial Engenharia Informática . Faculdade de Ciências Exatas e da Engenharia