Abstract
The aim of this research is to propose a methodology to build predictive models of academic performance through characteristics of engineering students in our country and to compare the models using different evaluation metrics. In this study, 228 students who are part of a public University in Mexico participated. Data were collected at the beginning of the course and, by means of three machine learning techniques, the predictive models were built. The characteristics of each model were analyzed and a prediction accuracy of around 65% was achieved. The model with the NaÁ¯ve Bayes technique was the most suitable for most of the metrics used in the study, mainly to identify students in danger of failure. In addition, it was found that the current average was the most significant characteristic for the prediction of the academic performance of the students participating in the study. The methodology developed can be replicated for other courses and the characteristics of the students can be collected at the beginning of the course or before, allowing the possibility of carrying out intervention strategies for students in danger of failure.