Student Performance Prediction Based on a Framework of Teacher’s Features

Teachers teaching skills are essential to motivate students’ engagement in online educational environments, where students and teachers interact with each other, generating a large amount of educational data. However, to the best of our knowledge, there is no previous study that takes advantage of the huge quantity of teachers’ behavioral data to predict students’ performance. To fill this research gap, we elaborated a theoretically based framework of teacher’s characteristics, that guided an automatic data collection of teachers’ behaviors to predict students’ performance. The implementation of a computational prediction system applied the Random Forest classifying algorithm, which achieved better performance, according to AUC metric, when compared to other algorithms. Two exploratory case studies were conducted to investigate the efficiency and efficacy of the framework of teacher’s features in Goiás Judicial School EJUG teachers in Brazil. The results from the case studies shown that the framework is effective to predict students’ performance. This work contributes to distant education, enabling monitoring teachers’ actions aiming students’ academic best achievements