Financial institutions use various machine learning techniques to create models to predict a customer's credit risk. In reality, customer behavior is constantly changing, so these models must be updated regularly. The objective of this project, based on a Kaggle competition, is to predict the probability of a client defaulting on their debt, considering the stability of the model over time. To do this, a database of more than 1,500,000 clients was used, taken from different sources in the banking sector in the United States.