Problems with late studies and desertion affect educational institutions, students who have them, and their families, hence the importance of studying them. In this work, we analyze the academic performance of the 2011-2016 cohorts of the Professional School of Systems Engineering of a public university, for this we have information from 976 students: university admission score, subjects’ qualifications and some personal data. The overall academic performance and the first year of study are statistically described. The adjusted weighted average and exogenous, endogenous, and total performance rates are calculated. With this set of variables, using a proprietary app and a commercial app, data mining techniques are applied to find patterns that describe student academic behavior. By applying classification techniques: Neural networks and decision trees, it is found that the most influential variables are the exogenous performance rate and the ratio of approved credits in relation to the credits that in theory had to be approved; for this the CRISP-DM methodology is used. As a result, some strategies are proposed that could decrease the studied problems. It is concluded that when analyzing a student’s academic performance, the qualifications obtained are not sufficient, their academic behavior, their performance in relation to their cohort and the pace of progress in the approval of the subjects should be considered. Thus, the techniques used allow students at academic risk to be identified in a timely manner.