Early Prediction of Student Profiles Based on Performance and Gaming Preferences

• Published in: IEEE transactions on learning technologies 2016-07, Vol.9 (3), p.272-284
• Main Authors: Barata, Gabriel, Gama, Sandra, Jorge, Joaquim, Goncalves, Daniel
• Format: Article
• Language: English
• Subjects: adaptive learning; Algorithms; Classification; Classification algorithms; cluster analysis; Cluster Grouping; Context; Education; Foreign Countries; Game Based Learning; Games; Gamified learning; Individual Differences; Learner Engagement; Learning; Multivariate Analysis; Prediction; Prediction algorithms; Preferences; Profiles; Silicon carbide; Student Characteristics; student classification; Student Records; Students
• ISSN: 1939-1382; 1939-1382; 2372-0050
• DOI: 10.1109/TLT.2016.2541664

State of the art research shows that gamified learning can be used to engage students and help them perform better. However, most studies use a one-size-fits-all approach to gamification, where individual differences and needs are ignored. In a previous study, we identified four types of students attending a gamified college course, characterized by different levels of performance, engagement and behavior. In this paper, we present a new experiment where we study what data best characterizes each of our student types and explore if this data can be used to predict a student's type early in the course. To this end, we used machine-learning algorithms to classify student data from one term and predict the students' type on another term. We identified two sets of relevant features that best describe our types, one containing only performance measurements and another also containing data regarding the students' gaming preferences. Results show that performance alone can be used to predict student type with 79 percent accuracy by midterm. However, its accuracy improves when paired with gaming data at earlier stages of the course. In this paper, we clearly describe our findings and discuss the lessons learned from this experiment.