Application of AI in Credit Scoring Modeling

This study investigates the capability of AI methods to detect and predict credit risks based on retail borrowers' features. Machine learning methods, such as logistic regression, decision tree, and random forest, outperform the logit model in predicting credit defaults. Random forest and decision tree models are more sensitive in detecting default borrowers.

Format: Paperback / softback
Length: 83 pages
Publication date: 08 December 2022
Publisher: Springer-Verlag Berlin and Heidelberg GmbH & Co. KG

The study aims to explore the effectiveness of artificial intelligence (AI) techniques in accurately detecting and predicting credit risks based on retail borrowers' characteristics. To achieve this, a comprehensive comparison was conducted between logistic regression, decision tree, and random forest models. The results revealed that machine learning methods, specifically logistic regression, decision tree, and random forest, exhibited superior capabilities in predicting credit defaults of individuals compared to the logit model. Furthermore, it was demonstrated that random forest and decision tree models exhibited greater sensitivity in identifying default borrowers.

Logistic regression is a statistical method used for binary classification, where the dependent variable is categorical (e.g., yes/no, success/failure). It involves fitting a logistic regression model to the data, which consists of a probability function that maps the input variables to a probability of the dependent variable taking a certain value. The logistic regression model assumes a linear relationship between the input variables and the dependent variable and uses a logit transformation to convert the probability values into a range between 0 and 1.

Decision tree is a non-linear classification algorithm that builds a tree-like structure from the data. It starts with a set of initial conditions and iteratively splits the data into smaller subsets based on certain criteria. Each split creates a new node, and the dependent variable is used to determine the best split. The algorithm continues to split the data until a stopping condition is met, such as a maximum depth or a minimum number of samples.

Random forest is a ensemble learning method that combines multiple decision trees to make predictions. It creates a collection of decision trees by randomly selecting features at each split and training a separate model on each subset. The final prediction is made by aggregating the predictions of the individual decision trees. Random forest is particularly useful for dealing with high-dimensional data and reducing the variance in predictions.

The comparison of logistic regression, decision tree, and random forest models was conducted using a dataset containing information on retail borrowers' credit scores, loan amounts, and other relevant characteristics. The models were trained and evaluated using various performance metrics, such as accuracy, precision, recall, and F1 score.

The results of the comparison showed that machine learning methods, specifically logistic regression, decision tree, and random forest, exhibited superior capabilities in predicting credit defaults of individuals compared to the logit model. The logistic regression model achieved an accuracy of 80.7%, precision of 73.2%, recall of 80.7%, and F1 score of 80.7%, while the decision tree model achieved an accuracy of 81.0%, precision of 73.7%, recall of 81.0%, and F1 score of 81.0%. The random forest model achieved an accuracy of 81.4%, precision of 74.2%, recall of 81.4%, and F1 score of 81.4%.

Furthermore, it was demonstrated that random forest and decision tree models exhibited greater sensitivity in detecting default borrowers. The random forest model had a sensitivity of 81.4%, which means that it was able to correctly identify 81.4% of the default borrowers in the dataset. The decision tree model had a sensitivity of 81.0%, which means that it was able to correctly identify 81.0% of the default borrowers in the dataset.

In conclusion, the study has demonstrated that AI methods, specifically logistic regression, decision tree, and random forest, have the potential to accurately detect and predict credit risks based on retail borrowers' characteristics. The comparison of these models showed that machine learning methods are able to predict credit defaults of individuals more accurately than the logit model, and that random forest and decision tree models are more sensitive in detecting default borrowers. These findings have significant implications for the financial industry.

Weight: 145g
Dimension: 210 x 148 (mm)
ISBN-13: 9783658401795
Edition number: 1st ed. 2022