The present work shows the analysis and modeling of a database with information about the various credit card transactions. The objective is to detect transactions that are fraudulent. In the modeling process, the �Ridge and Lasso�, �Boosting� and �Random Forest� techniques were applied in the modeling and variables selection. The results show that the accuracy of the models was very high, so the metric �Recall� was chosen as a second criterion for selecting the best model. This metric measures the percentage of positive values of the variable �fraud�. It is concluded that the best model is that of �Boosting� with 1,500 trees and a K-Folds of 10 that presented the best results in both training and validation. © 2020, Springer Nature Switzerland AG.