Predicting Student Success Using Data Generated in Traditional Educational Environments

Educational Data Mining (EDM) techniques offer unique opportunities to discover knowledge from data generated in educational environments. These techniques can assist tutors and researchers to predict future trends and behavior of students. This study examines the possibility of only using traditional, already available, course report data, generated over years by tutors, to apply EDM techniques. Based on five algorithms and two cross-validation methods we developed and evaluated five classification models in our experiments to identify the one with the best performance. A time segmentation approach and specific course performance attributes, collected in a classical manner from course reports, were used to determine students' performance. The models developed in this study can be used early in identifying students at risk and allow tutors to improve the academic performance of the students. By following the steps described in this paper other practitioners can revive their old data and use it to gain insight for their classes in the next academic year.