Design and Evaluation of Personalized Course Recommendations to Reduce Dropout Risk in Higher Education

Dropout rates in higher education remain high, particularly during students’ early semesters. Universities have become increasingly diverse, offering a broader range of degree programs. However, this diversity also brings challenges, particularly in transitioning from school to university. In Germany, up to 47% of dropouts occur within the first two semesters, with higher rates among male students, first-generation students, and those with a migration background. Academic difficulties, especially exam failure and perceived high-performance demands, are among the leading causes of dropout. This dissertation presents the design and evaluation of a course recommendation system aimed at supporting students at risk of dropping out. The system’s design is based on significant performance differences between students who dropped out and those who graduated, identified through data exploration from three undergraduate programs. Early involvement of students in the development process helped shape a transparent and interpretable recommendation system that is accessible to all students. The system uses an explainable nearest neighbors algorithm to recommend courses based on the academic performance of similar, successful students. Two success criteria are considered: (1) graduation and (2) passing a minimum number of courses in a semester. The second criterion enables earlier feedback on success and allows the system to adapt more quickly to curriculum changes. The system’s potential impact was comprehensively evaluated using historical data and multiple metrics, including overlap quality with passed courses, the number of recommended courses, and the change in predicted dropout risk. A key finding is that the number of recommended courses dynamically adapts to students’ academic performance, helping to prevent overload for at-risk students. The system also tends to recommend courses for at-risk students that are easier to pass, increasing their chances of passing and progressing academically. Course recommendations were more aligned with the courses passed by successful graduates, confirming the system’s potential to support academic success. A user study with 100 students further validated the system’s perceived explanation and recommendation quality. The results showed that students generally understand the explanations, although no clear preference emerged between list- and set-based presentation formats. This work contributes to the fields of Educational Data Mining and Learning Analytics by demonstrating how personalized course recommendations can support students at risk of dropping out while ensuring that successful students are not negatively affected. The findings highlight the importance of explainability and flexibility in course recommendation systems and provide insights for future research and practical implementation.