Patterns to Identify Dropout University Students with EducationalData Mining

This paper applies educational data mining algorithms to present an analysis of the most relevant characteristics of potential dropout students. The study used a dataset of 10,635 instances, acquired between 2014 and 2019 from 53 bachelor’s degree programs at a private university in the state of Puebla (Mexico). The results show that the model obtained from the decision trees performs better than other algorithms and allows for easy interpretation through decision rules. Furthermore, the model performs better than other related models in the literature that have been applied to the same problem. The methods used to select characteristics yielded the most important attributes to identify potential dropouts, such as the period, last semester completed, credits completed, attendance, courses failed, and program. These attributes and decision rules can be used to create mechanisms that help prevent dropout.