Recommender systems in higher education: The effectiveness of meta-data analysis in predicting academic success

A recommender system is: “a software technology that proactively suggests items of interest to users based on their objective behavior or their explicitly stated preferences” (Pu et al., 2011, p. 157). Whether we realize it or not, much of our lives are influenced by recommender systems. These systems may recommend where we eat, what movie we watch, what music we listen to, what products we buy, or even what news and social media content we see. Recommender systems have been effectively utilized in higher education to recommend courses as well as customizing content within an online course to meet specific student needs. The large volumes of data produced by e-learning systems, combined with other student and course data have been utilized with traditional statistical analysis as well other artificial intelligence techniques to predict student success. Utilizing course participation data, and other data collected from the e-learning allows for universities to provide academic support for students, but often the at-risk students are not identified until after the course begins. The purpose of this study was to develop and evaluate the effectiveness of content filtering and collaborative filtering recommender systems, in the grade prediction for university students. The information provided to the systems was limited to historical grade information. These recommenders were trained with over 377,000 individual course grades, from four years of university courses. The recommenders were then used to predict approximately 36,000 individual grades, with the predictions compared with the actual grades the students achieved. The collaborative filtering recommender system successfully predicted within a half grade 61% of the time. The recommender also was able to correctly predict 11% of the D and F grades. The results of the content filtering recommender system demonstrated that it was no more effective than simply predicting the average grade of all students in the course. This recommender successfully predicted 46% of the grades within a half grade. It was ultimately only successful in predicting less than 1% of the D and F grades correctly. The study demonstrated the potential for recommender systems to be utilized in grade prediction and early warning for students that may have trouble in a course, it opens several paths for future stud