Learning Profiles to Assess Educational Prediction Systems

Distance learning institutions record a high failure and dropout rate every year. This phenomenon is due to several reasons such as the total autonomy of learners and the lack of regular monitoring. Therefore, education stakeholders need a system which enables them the prediction of at-risk learners. This solution is commonly adopted in the state of the art. However, its evaluation is not generic and does not take into account the diversity of learners. In this paper, we propose a complete methodology which objective is a more detailed evaluation of a proposed educational prediction system. This process aims to ensure good performances of the system, regardless of the learning profiles. The proposed methodology combines both the identification of personas existing in a learning context and the evaluation of a prediction system according to it. To meet this challenge, we used a real dataset of k-12 learners enrolled in a french distance education institution.