Detecting carelessness through contextual estimation of slip probabilities among students using an intelligent tutor for mathematics

Abstract

A student is said to have committed a careless error when a student's answer is wrong despite the fact that he or she knows the answer (Clements, 1982). In this paper, educational data mining techniques are used to analyze log files produced by a cognitive tutor for Scatterplots to derive a model and detector for carelessness. Bayesian Knowledge Tracing and its variant, the Contextual-Slip-and-Guess Estimation, are used to model and predict carelessness behavior in the Scatterplot Tutor. The study examines as well the robustness of this detector to a major difference in the tutor's interface, namely the presence or absence of an embodied conversational agent, as well as robustness to data from a different school setting (USA versus Philippines). © 2011 Springer-Verlag Berlin Heidelberg.