Data-driven modelling of students' interactions in an ILE

This paper presents the development of two related machine-learned models which predict (a) whether a student can answer correctly questions in an ILE without requesting help and (b) whether a student’s interaction is beneficial in terms of learning. After presenting the rationale behind the development of these models, the paper discusses how the data collection was facilitated by the integration of different versions of the ILE in realistic classroom situations. The main focus of the paper is the use of the ICS algorithm of WEKA to derive Bayesian networks which provide satisfactory predictions. The results are compared against decision trees and logistic regression. The application of these models in the ILE and how their potential educational consequences were taken into account are outlined followed by a discussion of future lines of research.