Introduction:

University dropout is now considered a complex phenomenon that goes beyond the number of students not enrolled and that is continuously growing, especially in the first years of study.

Objective:

In the present study, a prediction model combining Survival Analysis, Decision Trees, and Random Forest, under the Machine Learning philosophy, is proposed for the early diagnosis of possible factors causing dropout in university students.

Methods:

The proposal consists of 3 phases: the Survival Analysis that allows estimating the probability of permanence of the student (survival). Phase 2 starts from the probability value obtained in the previous phase and uses it as a response variable in the modeling process based on Decision Trees to establish survival patterns around the variables considered. Finally, in phase 3, the critical variables in the model are identified using the Random Forest.

Results:

The proposed methodology allowed the design of a prediction model to identify the main segmentation variables in behavior patterns of possible cases of academic dropout.

Conclusion:

Even though the proposal was developed considering a particular case of a Chilean university, the efficient combination of metaheuristics allows the extrapolation of the methodology to any context and academic reality. However, the conditions and needs of each institution must be considered.

        · abstract in Portuguese | Spanish     · text in English     · English ( pdf )