School dropout is a complex phenomenon that affects the socioeconomic performance of a country and for many decades has been the object of study by researchers from different areas around the world. Having an interdisciplinary character, it is observed that studies on the phenomenon of dropout have made use of quantitative analytical models, resorting in particular to the use of data analysis methodologies. Therefore, this dissertation, inserted in the field of
Digital Humanities, aims to research Data Analytics approaches to support Higher Education Policies in undergraduate courses (bachelors and licentiates) in the face-to-face teaching modality, specifically aimed at the control and fight against school dropout. The research conducted addressed two fronts: (a) systematic literature review – where, as the name implies, search criteria are used systematically to collect, identify and select relevant scientific works from the literature relevant to the topic; and (b) creation of a methodology based on Design Science Research to develop the analysis of works in the literature. The proposed methodology is composed of four components: Framework, Theorization, Modeling and Experimental Protocol. The protocol developed to guide the Systematic Review was satisfactory, returning 42 articles for analysis. In the Framing analysis, it was found that the most used Data Analytics task is the predictive one, and among these, there was a predominance in the use of individual techniques to the detriment of the ensembles methods, with the Decision Tree being one of the
most used. Less than 50% of studies define the term dropout and 70% of them treat this phenomenon as a classification task. Regarding theorization, academic information is the most considered for the construction of models. Much of the work starts from the theory that academic performance is an important predictor of dropout. As for Modeling, it was evaluated that most studies use only one set of data, whose origin can be from information from the academic system (internal sources); institutional surveys, which comprise national, regional and academic databases or questionnaires used by the researchers themselves to acquire more specific information (external sources). In addition, a combination of information (demographic background, previous school performance/information and academic information/performance) was used to build the models. In the analysis of the Experimental Protocol, it was observed that the most used model adjustment method was cross-validation and the most used metric of interest was Accuracy, present in 26 studies. These results and analyzes led to the construction of a mental map, organizing the main proposals in the literature. The methodology proposed based on the DSR was fundamental for the analysis of the works, allowing the identification of Data Analytics approaches present in the investigated works in an orthogonal way to the components, contributing to future research to benefit from this methodology, especially regarding the creation of computational artifacts. The study also showed that it is possible to use the Data Analytics approach to deal with student dropout, eventually helping to mitigate the effects of this phenomenon in Higher Education.