A Predictive Model of Academic Failure or Success for Institutional and Trace Data

Abstract

In recent years, Educational Data Mining (EDM) has contributed significantly to the prevention of various challenges in many sectors including the academia. This paper brings a comparative study of EDM impact, using institutional data and trace data generated by a virtual learning environment to predict academic failure or success. For this, we created a model of educational data mining using logistic regression as a classifier and cross validation as a test method. Logistic Regression is widely used as a classifier (or predictor) in educational data mining. The developed model was applied to a public data set of 32,593 students, distributed among seven courses. The results show that the application of the proposed model in the trace data favors excellent predictions, on average 42% better, in relation to using just the institutional data. The results also revealed that it is better to separate the trace data from the institutional data in the model application, since the gain in the prediction is negligible and the computational time spent is around four times greater if the both data sets are combined. In addition to contributing to the findings, we also believe that our methodology can be re-applied by researchers, educators and education managers.