EDM classifiers performance and comprehensibility evaluation

Selecting a classification model to deploy in educational systems is a very sensitive process since it is necessary to guarantee that the model will work well after being deployed in a real situation and at the same time the model needs to be friendly to the users involved in the educational settings. The goal of this paper is to select an accurate and easy-to-understand model to serve as a basis to build a new reporting module in Moodle system for predicting students' success. With this objective, two experiments were conducted in order to assess the quality of different classifiers with respect to performance and comprehensibility. In the first experiment, the performances of different models were evaluated based on the Accuracy criteria. In the second experiment, different respondents provided their subjective opinion about the models' comprehensibility. The results discovered that the PART model was the most suited model to implement the report for predicting student's success.