Staying Ahead of the Curve: Selecting Students for Newly Arising Tasks.

Adaptive instructional systems (AISs) usually involve some form of task selection. By selecting certain tasks for a student, the AIS aims to improve the student’s skills and performance. In this paper, we explore the opposite of this concept, namely student selection : given (a) the past performance of a set of students on a set of tasks, and (b) a newly arising (and perhaps previously unseen) task, we aim to select the student that is expected to reach the best performance on this task. We investigate three methods for selecting students: (1) matrix factorization and (2) a neighborhood model, both collaborative filtering methods commonly found in recommender systems, and (3) random forest regression, a content-based filtering method that aims to predict the exact score of a given student on a certain task. For a proof-of-concept, we construct a data set of the performance data of various machine learning algorithms (i.e., virtual students) on a set of video games (i.e., virtual tasks), and apply the three methods to this data set. We present the results of the application, and then conclude the paper by discussing the potential and the limitations of our research.