The objective of this paper is to develop a predictive model using machine learning to determine customers' payment behavior with data from an Ecuadorian Bank. This predictive model can help the bank anticipate the behavior of customers who are likely to default or stop paying a loan so that appropriate measures can be taken, such as selling the portfolio or putting pressure on the collection. The analysis of customer payment behavior was performed using time series, and four machine learning models were used Recurrent Neural Networks Model, Convolutional Temporal Networks Model, TRANSFORMER Model and N-BEATS Model using PYTHON's DARTS library. These models were compared with the classical Markov chains model, and three metrics were used to compare them: i) Error, ii) F-Score, iii) Confusion Matrix and their behavior was analyzed by increasing the number of predictions made. The final part analyzes the reasons for the models not being so robust and proposes how machine learning models could be better applied for this type of prediction.

Machine Learning, Markov, Time Series, Banking