Aplicação de métodos de mineração de dados na análise da evasão escolar em cursos STEM da UTFPR

Abstract

Higher education institutions in Brazil have high dropout rates, a matter of concern because dropout leads to reduced resources and the waste of public investments, in addition to compromising students’ academic and professional prospects. In this context, the analysis of university data, which are increasingly available, becomes essential for understanding this phenomenon and developing retention strategies. Seeking to contribute to this scenario, this study employed data mining techniques, combined with quantitative and qualitative analysis, regarding students who dropped out of undergraduate courses in the fields of Science, Technology, Engineering, and Mathematics at the Federal University of Technology – Paraná (UTFPR). To this end, two complementary data sources were used: a questionnaire answered by 344 students who had dropped out of UTFPR, Pato Branco campus, and data from UTFPR’s Analytical Management Reports (RAG). The results from the first data source indicated a predominance of personal reasons among the factors associated with dropout, followed by academic and financial reasons. In the data mining stage, using the second data source, classification algorithms were tested in the WEKA software and through the Python programming language, with Random Forest and J48 showing the highest accuracy and precision among the evaluated models. Among the analyzed attributes, the academic period attended and the academic performance coefficient stood out as the most relevant for classifying records between dropout and non-dropout students. Based on these findings, this study is expected to contribute to university management in the development of action plans aimed at mitigating dropout rates and promoting student retention.