In higher education, students must select a set of courses every semester. Academic advising is crucial to their decision-making. The main objective of this dissertation is to provide novel techniques to build efficient and accurate course recommendation systems using machine learning to assist student advising and address other related problems in educational settings. First, using the course registration history of past students, we develop three session-based course recommendation (CR) models, two based on deep neural networks (CourseBEACON and CourseDREAM) and one on tensor factorization (TF-CoC). These models capture the association and relationship of courses taken together within a session (semester) to recommend a set of well-suited courses for the upcoming semester personalized to each student. Second, we re-purpose the CR models to estimate how many students will be enrolled in each course offered in the next semester. A course enrollment prediction tool is helpful to the administrators of a department who prepare course offerings every semester. Third, we explore the use of large language models (LLMs) for course recommendation. We investigate two different ways to utilize LLMs indirectly and directly (offline, to ensure no sensitive data leakage) to generate recommendations from pre-trained or fine-tuned LLMs. The next goal of this thesis will be to explore how these course recommendations can be more useful for students and advisors with explainability. Our plan is to investigate interpretable machine learning models as well as the power and capabilities of LLMs towards our goal. By enhancing transparency, we can better inform student advising, decision making, and students’ success.

Md Akib Zabed Khan is a Ph.D. candidate at the Knight Foundation School of Computing and Information Sciences at Florida International University, working under the supervision of Dr. Agoritsa Polyzou. Before starting his Ph.D., he received M.Sc. and B.Sc. degrees in Computer Science and Engineering from Jahangirnagar University, Bangladesh in 2019 and 2018, respectively. He also served as a Lecturer at two different private universities in Bangladesh for 3 years (2019-2021). His work spans leveraging different data mining techniques, deep learning, and large language models to develop recommendation systems and predictive tools for the education sector. To date, he has 9 publications, including two conference papers in the most relevant venue, the International Conference on Educational Data Mining (EDM2023, EDM2024), and one journal paper at the Journal of EDM (JEDM).