Evaluating the impact of prior required scaffolding items on the improvement of student performance prediction

Recently, tracking student behavior has become a very important phase for constructing adaptive educational systems. Several researchers have developed various methods based on machine learning for better tracing students' knowledge. Most of these methods have shown an effective estimation of student features and an accurate prediction of future performance. However, these methods recognized certain limitations since they use only the correctness of prior student responses to make predictions without paying attention to many other important student behaviors. In addition, researchers have only considered scaffolding items as a pure method of learning without having analyzed student performance at the time of answering these items. Our purpose in this article is to conduct an experiment that aims to evaluate how best to use data about the prior required scaffolding items to predict future student performance. For this reason, we proposed two separate models, namely, the first one identifies whether a student has previously required to use scaffolding items prior main question or has immediately answered it without requiring assistance. For the second model, as an improvement of model 1, our objective is to improve the student's performance under the constraint of answering scaffolding items. The performance of our two models is evaluated against the original Performance Factors Analysis algorithm to mark differences. The results show that the two proposed models provide a positive improvement in predicting the future performance of students. Moreover, our second model can reliably increase the predictive accuracy.