Course Prophet: A System for Predicting Course Failures with Machine Learning: A Numerical Methods Case Study

In this study, our purpose was to conceptualize a machine-learning-driven system capable of predicting whether a given student is at risk of failing a course, relying exclusively on their performance in prerequisite courses. Our research centers around students pursuing a bachelor's degree in systems engineering at the University of Cordoba, Colombia. Specifically, we concentrate on the predictive task of identifying students who are at risk of failing the numerical methods course. To achieve this goal, we collected a dataset sourced from the academic histories of 103 students, encompassing both those who failed and those who successfully passed the aforementioned course. We used this dataset to conduct an empirical study to evaluate various machine learning methods. The results of this study revealed that the Gaussian process with Matern kernel outperformed the other methods we studied. This particular method attained the highest accuracy (80.45%), demonstrating a favorable trade-off between precision and recall. The harmonic mean of precision and recall stood at 72.52%. As far as we know, prior research utilizing a similar vector representation of students' academic histories, as employed in our study, had not achieved this level of prediction accuracy. In conclusion, the main contribution of this research is the inception of the prototype named Course Prophet. Leveraging the Gaussian process, this tool adeptly identifies students who face a higher probability of encountering challenges in the numerical methods course, based on their performance in prerequisite courses.