A Machine Learning-Based Course Enrollment Recommender System
Location
SU 214
Department
Computer Science
Abstract
At Northeastern Illinois University, the Computer Science (CS) department offers students a wide range of unique courses. These courses reflect the diversity of options available at the university. Students can choose from three concentrations within the CS major. Nevertheless, students pursuing a degree in computer science will not always be enrolled in the same courses. They often have difficulties choosing which courses to enroll in as part of their concentration. This research proposes a personalized course recommender system to help current and future students find the courses they need to enroll in. It will generate a list of suggested courses for each student to consider registering for the upcoming semester. The system utilizes a collaborative filtering algorithm to profile each student with respect to their registration preferences. The algorithm extracts latent features of students and courses by performing stochastic gradient descent to optimize our objective function. The learning process takes into account auxiliary information, such as course instructors, meeting times, and delivery methods (online/in-person/hybrid). When making recommendations, the filtering step of the system follows program requirements to refine the output by removing unnecessary courses and substituting courses with their prerequisites when necessary. We evaluated our proposed model on a CS enrollment dataset. In the experiments, we conducted an extensive study on the hyperparameters of the model and visualized each parameter. The results show that our model can produce high-quality recommendations, whose accuracy is comparable to state-of-the-art research. A web application that demonstrates the framework is also implemented.
Faculty Sponsor
Xiwei Wang, Northeastern Illinois University
A Machine Learning-Based Course Enrollment Recommender System
SU 214
At Northeastern Illinois University, the Computer Science (CS) department offers students a wide range of unique courses. These courses reflect the diversity of options available at the university. Students can choose from three concentrations within the CS major. Nevertheless, students pursuing a degree in computer science will not always be enrolled in the same courses. They often have difficulties choosing which courses to enroll in as part of their concentration. This research proposes a personalized course recommender system to help current and future students find the courses they need to enroll in. It will generate a list of suggested courses for each student to consider registering for the upcoming semester. The system utilizes a collaborative filtering algorithm to profile each student with respect to their registration preferences. The algorithm extracts latent features of students and courses by performing stochastic gradient descent to optimize our objective function. The learning process takes into account auxiliary information, such as course instructors, meeting times, and delivery methods (online/in-person/hybrid). When making recommendations, the filtering step of the system follows program requirements to refine the output by removing unnecessary courses and substituting courses with their prerequisites when necessary. We evaluated our proposed model on a CS enrollment dataset. In the experiments, we conducted an extensive study on the hyperparameters of the model and visualized each parameter. The results show that our model can produce high-quality recommendations, whose accuracy is comparable to state-of-the-art research. A web application that demonstrates the framework is also implemented.