Predicting student performance with data from an interactive learning system

Abstract

Nowadays Interactive Learning Systems have been developed to provide students with new forms of practicing concepts. In this work we propose to predict if the student fails or succeeds in the introductory mathematics course based on the information collected by an interactive learning platform. The predicting models are based on binary support vector machines (SVM). As some of the collected data sets are unbalanced the study was conducted with suitable strategies to train this binary classifier.SIACUA - Sistema Interativo de Aprendizagem por Computador, Universidade de Aveiro - is a web application designed to support autonomous study. For each subject is defined a concept map with questions associated to each concept. The system is supplied with parametrized questions from PmatE (pmate.ua.pt) and MEGUA (cms.ua.pt/megua) projects. It implements a user model based on Bayesian networks.