An Exploratory Study on the Use of Machine Learning to Predict Student Academic Performance

Literature Review

Educational Data Mining (EDM) has been defined as “…scientific inquiry focused on developments of methods for making discoveries with data from educational settings and using those methods to better understand students and the settings which they learn in…” (Baker, 2010). Understanding students through educational data mining can give new insights to ways that can improve student academic performance. Academic success is seen as a critical factor for individual success in contemporary society (Pritchard & Wilson, 2003). If students’ academic performance can be previously predicted, it gives policy makers the opportunity to introduce policies that will improve student academic success rate, thereby increasing the likelihood of successful completion of a higher education degree. Also, creating predictive models that can be used for early identification of weak students who will be at risk is beneficial for reducing failure or dropout rates in higher education institutions (Raju & Schumacker, 2016).

Techniques for predicting student performance have been researched extensively. Historically, correlation and multiple regression are the traditional methods used to investigate the extent to which socioeconomic or psychological factors can positively or negatively affect a student’s academic performance. For example, Figilo and Kenny (2007) have used correlation to investigate the relationship between teacher incentives and student performance. However, machine learning techniques can use these socioeconomic, psychological or other factors to also predict how well a student will perform in a particular subject, thus it is a useful technique that can be used in the design of systems that can provide real time guidance to students.