Models and Algorithms for Performance Prediction and Course Recommendation in Higher Education

Thumbnail Image

Persistent link to this item

View Statistics

Journal Title

Journal ISSN

Volume Title


Models and Algorithms for Performance Prediction and Course Recommendation in Higher Education

Published Date




Thesis or Dissertation


Educational institutions need to use supporting tools in order to reduce high student drop-out rates and ensure their students' timely graduation. Educational data mining involves the development of such methods that leverage student data. Their purpose is to generate warnings about students that are at risk of failing, identify the enrollment practices that are associated with academic success, and allow for suitable interventions. The application of these models in a higher education institution can improve advising and degree planning. The objective of this thesis is to develop methods that will support students to make informed decisions regarding their course registration in the context of higher education. We also explore fairness concerns that might arise in a course recommendation system. We want to create models that can be used before the semester starts in order to allow the students to make any necessary adjustments in their plans. Instructors can also benefit from them, as it will allow them to shape their expectations about the enrolled students in their class, and adjust the syllabus and course material accordingly. For the purpose of this work, we will focus on traditional higher education datasets and develop models that do not need any information for the structure and the syllabus of a particular course. All we use is the students' transcript data. In this setting, we first introduce next-term grade prediction methods to estimate the grades that a student is expected to receive for the courses they plan on taking the following semester. These algorithms are based on sparse linear and low-rank matrix factorization models that are specific to each course or student-course tuple. These methods identify the predictive subsets of prior courses on a course-by-course basis. Second, we delve into the factors that influence students' performance. We extract informative features from the students' transcript data and form the problem as a binary classification one, to identify the students that have poor performance. We consider a student's performance both in the absolute sense (in terms of grades achieved), but also in the relative sense (compared to the student's past performance). We present a comprehensive study to answer the following questions: which features are good indicators of a student’s performance? which features are the most important? The findings are interesting, as different features are the most important for different classification tasks. Next, we addressed the problem of course recommendation. Recommender systems have been extensively used in various domains to support decision making and empower user choices. Within the educational domain, recommenders can generate a personalized list of courses for each student to consider taking in the next semester. Towards that direction, we propose Scholars Walk, a random-walk-based approach that captures the sequential relationships between the different courses. Based on the ''wisdom of the crowd" and the students' prior courses, we recommend a shortlist of courses for next semester. Our framework is very efficient, easily interpretable, while also being able to take into consideration important aspects of the educational domain. Finally, we explore fairness in the context of course recommendation. It is important to ensure that the models that we develop and apply in an educational institution do not discriminate against particular groups of students. All students should be treated fairly and offered similar opportunities and quality of services based on the notion of group fairness. Based on this idea, we examine the fairness of the opportunities offered to the students by the system's recommendations. We formulate our approach as a multi-objective optimization problem, and we study the trade-offs between equal opportunity and quality. The algorithm first makes an initial assignment of courses to recommend to the students, which we assume is of optimal quality. The model then refines the initial solution to improve overall fairness. The experimental evaluation with synthetic and real datasets showcase that we can promote equality of opportunity in the recommendations without significantly weakening their quality.



University of Minnesota Ph.D. dissertation.August 2020. Major: Computer Science. Advisor: George Karypis. 1 computer file (PDF); xi, 93 pages.

Related to




Series/Report Number

Funding information

Isbn identifier

Doi identifier

Previously Published Citation

Suggested citation

Polyzou, Agoritsa. (2020). Models and Algorithms for Performance Prediction and Course Recommendation in Higher Education. Retrieved from the University Digital Conservancy,

Content distributed via the University Digital Conservancy may be subject to additional license and use restrictions applied by the depositor. By using these files, users agree to the Terms of Use. Materials in the UDC may contain content that is disturbing and/or harmful. For more information, please see our statement on harmful content in digital repositories.