Early Identification of Struggling Students in Large Computer Science Courses: A Replication Study

The identification of struggling students in a scalable manner in large computer science courses continues to attract researchers' attention because of the high failure and dropout rates in such courses. In the last two decades, studies on this topic made significant progress in taking advantage of dynamic data sources and employing machine learning techniques to identify struggling students. However, the success of these studies was still limited due to reasons such as oversimplification and utilizing exclusive tools. These limitations make it difficult to replicate the findings of prior research on this topic. To address these issues and explore the extent to which we can replicate studies on this topic, this study replicated a recent study that explored identifying struggling students at the topic level using context-agnostic features. Our results demonstrate the potential of context-agnostic features in identifying struggling students with varied success rates. Our findings shed light on the robustness and feasibility of using machine learning techniques to identify struggling students in large computer science courses. Finally, our discussion provides useful guidance on future studies and replications on this topic.