Predicting Student Engagement in the Online Learning Environment

Introduction

In earlier times, educational opportunities have been limited by the resources within schools. Technology-enabled learning allows learners to access resources anywhere in the world (U.S. Department of Education, 2017). The online learning is sought by those who want to pursue their education while accomplishing the other responsibilities of life such as work and family besides the learning. These students who have burdens of many responsibilities face attrition (Dixson, 2015).

Engagement is defined as “the behavioural intensity and emotional quality of a person's active involvement during a task” (Reeve et al., 2004, p.1). Manwaring et al. (2017) studied engagement at three distinct levels of analysis: the institutional level, the course level, and the activity level. “Activity level engagement has received less attention and research than institutional and course level engagement” (Manwaring et al., 2017, p.2). Student engagement in online learning requires advance study as the online existence of universities has improved. Engagement in the online learning environment never obtained due consideration in the past. (Dixon, 2015).

Moreover, if a student loses interest or is not getting engaged in the e-learning session, the teacher cannot easily monitor as the setting is online learning (Al-Alwani, 2016). And because engagement represents a direct pathway to learning, disengagement (losing interest or not getting engaged) provides barriers to achieving learning outcomes (Hancock and Zubrick, 2015).

Technology-mediated learning provides significant student engagement data which is unavailable in more traditional contexts. Many of the systems used in technology-mediated learning keep records of real-time data about student interactions with the system (Henrie et al., 2015). Husain et al., (2018) applied supervised machine learning algorithms to predict low student engagement from interaction in virtual learning environments. Motz et al., (2019) applied logistic regression model through a clustering technique to predict student engagement from interaction in learning management system which is Canvas. Cocea and Weibelzahl (2011) developed a disengagement prediction model on data of an e-learning system called HTML tutor. In these works, the models predicted student engagement from behavioural factors alone. However, engagement needs to be defined as multi factor construct to ensure that the richness of real human experience is understood (Henrie et al., 2015). Sadeque et al., (2015) developed logistic regression model to predict continued participation in an online health forum. However, the features of the discussion forum occurred in health related discussion, not e-learning related. Moreover, the authors discussed that relations between features such as number of replies to someone’s post and the time between someone’s post and replies he/she got and engagement are unknown. Sharma et al., (2019) also predicted student engagement from emotional factors with facial emotion recognition tools. Moreover, Calvo and D'Mello, (2010) remarked that affect detection systems that integrate data from different factors have been widely advocated but rarely implemented. Kizilcec et al.,(2013) also pointed out that constructing a model of engagement with smallest granule of time has not been implemented widely, but implementing it is important as it allows to uncover more subtle patterns. There are two research questions in this study. These are: