Towards a More Accurate Knowledge Level Estimation

Adaptive testing or Computer Adaptive Testing (CAT), adjusts the set of questions to the learnerpsilas estimated ability level, in order to reduce the guessing or slipping in answering too difficult or too easy questions. But the accuracy and fairness of this kind of assessment can be questioned. In this paper we presented a new approach for adapted assessment, in which the fixed set of questions is scored adaptively to maintain the benefits of adaptive testing while improving the fairness and accuracy. A dedicated Bayesian Network model is used to address the guessing and slipping, according to the estimated learnerpsilas level of knowledge and the difficulty level of the questions. The proposed system was implemented and used to estimate studentspsila level of knowledge in a virtual course. Then its accuracy was tested by predicting scores of students in a conventional exam. The results are promising and encourage further development of this method.