Mineração de dados educacionais (MDE) com uso de algoritmos de machine learning para análise da evasão estudantil numa instituição de ensino público federal do estado da Paraíba

Abstract

Student dropout has been a recurring theme in Educational Data Mining and a global concern, especially due to its socio-economic impacts that bring quite negative consequences to any education system, involving a multitude of factors, both internal and external to these school environments. In this line of research, the present study aims to identify the variables associated with this phenomenon in a higher education public institution located in the state of Paraíba. To do so, the development of a model will be proposed using the Ensemble Machine Learning algorithm, opting for Random Forests. Data will be collected from the academic records systems of the institution under study, selecting the data of administration students as a pilot. The methodology adopted follows the CRISP-DM standard, which demonstrates, sequentially and incrementally, the evolution of the stages of the proposed model construction. As a result, the model demonstrated its predictive power with an accuracy of 81%, helping the institution’s managers to identify the most significant variables for the existence of the problem and, thus, promote better-directed collaborative dialogues and actions among the multi-professional teams in an attempt to contain the problem.