Modeling student retention in an environment with delayed testing

Student modeling has been widely used in the prediction of student correctness behavior on the immediate next action. Some researchers have been working on student modeling to predict delayed performance, that is, retention. Prior work has found that the factors influencing retention differ from those that influence short-term performance. However, this prior research did not use data which were specially targeted to measure retention. In this study, we describe our experiments of using dedicated retention performance data to test the students’ ability to retain, and experiment with a new feature called mastery speed, indicates how many problems the students need to attain initial mastery. We found that this new feature is the most useful of our features. It’s not only a helpful predictor for 7-day retention tests, but also a long-term factor that influences students’ later retention tests even after 105 days. We also found that, although statistically reliable, most features are not useful predictors, such as the number of students’ previous correct and incorrect responses which are not as helpful in predicting students’ retention performance as in PFA.