Abstract

ESPINOSA-ZUNIGA, Javier Jesús. Application of Random Forest and XGBoost algorithms based on a credit card applications database. Ing. invest. y tecnol. [online]. 2020, vol.21, n.3, 00002.  Epub Dec 02, 2020. ISSN 2594-0732.  https://doi.org/10.22201/fi.25940732e.2020.21.3.022.

Two of the existing machine learning algorithms currently stand out: Random Forest and XGBoost. Both have become very popular. Random Forest is an algorithm that emerged almost twenty years ago and is widely used for the balance it offers between complexity and results. On the other hand, XGBoost is an algorithm that has aroused great interest because although it is relatively recent, it is currently considered the state of the art in machine learning algorithms for its results. One of the sectors in which this type of algorithm is applied is the financial. Some examples of its application in this sector are: customer segmentation, fraud detection, sales forecasting, customer authentication and market behavior analysis. An area of particular interest in this sector is the identification of clients to whom to grant a credit card: this is critical for financial institutions since an incorrect selection of these clients could lead to an increase in their past due portfolio. In the present study the Random Forest and XGBoost algorithms were applied on a credit card application database (donated by an Australian bank for research purposes) to identify the applications most likely to be granted a credit card. The models obtained were compared statistically (from which the model obtained with the XGBoost algorithm was selected) and the results were presented with graphs that allow answering two key questions from the business perspective: what are the requests to which a card must be awarded? and what results do we expect if the model is applied? The most important contribution of the present study is to apply two very effective algorithms on this database with a business focus.

Keywords : Machine Learning; XGBoost; Random Forest; decision tree; hyper parameter.