Redesigning Prediction Algorithms for At-Risk Students in Higher Education: The Opportunities and Challenges of Using Classification Techniques in a University Academic Writing

Introduction

Industry 4.0 is a term that “designates the increasing interlinking of production and information as well as communications technology” (Graube & Mammes, 2017, p. 844). With the emergence of Industry 4.0, there is a pressing need for higher education faculty members to develop algorithms that fit into changing need of intelligent systems.

At the university level, faculty members in higher education have started using new technologies that have emerged in the context of Industry 4.0. Some have started with a simple motion-detected lighting system in their classrooms, or a timed air conditioning system in their office. Others have started with more adaptive learning management systems that can cater to the needs of students at different knowledge levels (Ciolacu & Beer, 2016). These systems are only starting points, and most faculty members are aware of the greater scope of emerging Industry 4.0 technologies for use in higher education.

There are course-level opportunities for faculty members to make use of Industry 4.0 technologies; establishing algorithms to identify at-risk students is one of them. In the previous generation, the use of mass lecture was prevalent in higher education and this happens in all classrooms, including a language classroom (Abdullah, Ramli, & Rafek, 2017). Due to the high number of students in mass lectures, it is hard for lecturers to give sufficient feedback and attention to individual students (Chingos & Whitehurst, 2011). It is possible that students enjoy having a good chat with their lecturers and that they may be facing problems in their studies. Without the support of computers or algorithms, these students may end up failing the course, with lecturers only realizing problems at the end of the term. This gives rise to an emerging line of research, “actionable intelligence” which includes identifying students who are struggling in a course. It also includes predicting students’ academic success (Park, Yu, & Jo, 2016).

A great range of studies have outlined the benefits of using learning analytics to identify at-risk students. For example, Klüsener and Fortenbacher (2015) and Bainbridge et al. (2018) highlighted the need for learning analytics to predict at-risk students in an online environment. Lawson, Beer, Rossi, Moore, and Fleming (2016) reported on a university-wide early prediction system. Essa and Ayad (2012) built a similar system and concluded that the early warning system is a bridge between algorithms and support interventions. Several researchers believe that using algorithms to predict students’ performance will be common and useful in the future, when there are more online (or blended learning) courses and when data are available for exploration. It is therefore time for faculty members in Industry 4.0 to take this one step forward and identify and support students who are at-risk.

With emergence of the trend for predicting which students may be at risk, more applications are available (Clow, 2013; Wolff, Zdrahal, Nikolov, & Pantucek, 2013), but not all are useful. One of the leading learning management providers, Blackboard, offers a tool called Retention Center that derives algorithms and rules to help identify at-risk students. A student is marked at-risk if he or she misses a deadline or has scores that are lower than average (Blackboard, 2018). The algorithms, which Blackboard calls rules, seem to overlook the differences among students across cohorts and place unnecessary emphasis on certain incidental factors (e.g. submitting the assignment late by a few hours). These flaws may affect accuracy in predicting at-risk students.