Cluster-based Outlier Analysis of Carefulness Among Students using Physics Playground

Abstract

We explore a student carefulness model using cluster-based outlier analysis. In a related work by the authors, a predictive model of student carefulness was created, built and empirically validated using Philippine samples. Carefulness was found to exist in the dataset and could be robustly predicted using features derived from Physics Playground’s interaction logs. In this work, we found that clusters of outliers existed in the dataset and studied how these affect the model. In our prior work we found that carefulness did not have any linear relationship with post-test learning gains. Investigating the outliers, in this study, resulted to findings that post-test learning gains of the outlying (least careful) and non-outlying (more careful) groups are significantly different. Further, we also found that the degrees of carefulness between the clusters within the non-outlying groups were also significantly different. With this finding, educational pedagogies and interventions can be more effective when we consider that carefulness among students are varied and can be addressed distinctly and not in general to be able to achieve the desired learning gains.