2020-11-27, 15:50–16:10 (Europe/Athens), Room 1
Predicting students’ learning outcomes is one of the main topics of interest in the area of Educational Data Mining (EDM) and Learning Analytics (LA). A great number of research studies have been focused on the development and deployment of machine learning algorithms for solving a variety of predictive problems, such as detecting whether a student is going to pass or fail a certain course. The accurate identification of students at risk of fail is of utmost importance for universities, since pedagogical assistance and well-planned interventions could be offered to enhance student learning, improve their performance and avoid failure.
To this end, we examined the available educational data that we currently store in our premises. We mainly focused on data that is collected from two sources; our e-learning Moodle platform and our local student information system. We selected 9 blended semester courses that were held during the academic years 2017-2018, and 2018-2019. Each course was supported by our e-learning Moodle platform, with several learning resources and activities enabled; forums, pages, recourses, folders, urls, assignments, quiz, workshops. In a testing environment, we developed a custom report plugin for Moodle that allowed us to export a collection of course datasets modeled for pass/fail classification tasks. The dataset of each course contained attributes related to students’ on line activity and performance (stored at our Moodle), along with demographic data and students’ previous academic achievements (stored at our SIS).
We carried out a plethora of experiments using state of the art machine learning algorithms (i.e. SVM, k-NN, Random Forests, XGBoost, and Deep NN) for predicting whether a student is going to pass of fail on each course. The results demonstrated satisfactory performance in most cases. At a next phase, we separated our course datasets into training and testing sets (7 train and 2 test courses). Using the best performer of our experiments and the training course datasets, we created a learning model. Afterwards, we expanded our custom Moodle plugin in order to print the prediction results for students our test courses. Hopefully our work will provide a useful contribution to the MoodleMoot community and will be a first step towards the application of EDM and LA in our educational settings.
Andreas Sapountzis, Vasiliki Kalfa, Dimitrios Pesios, Maria Tsiakmaki and Konstantinos Karaoglanoglou are currently working in the IT Center of the Aristotle University of Thessaloniki (AUTh).