To Complete or Not to Complete: Prediction and Classification of Dropouts in a Cs1 Course

Despite the growing demand for programming skills, introductory computer science courses (CS1) remain a significant hurdle for many students, often leading to high dropout rates. The considerable number of new concepts can be challenging for beginners, leading to misunderstandings and frustration within the learning process. Addressing the factors contributing to student dropout in CS1 is essential to improve retention rates. Therefore, this study analyzed the behavior of 747 students enrolled in a CS1 course at the University of Sydney during the second semester of 2023. Data from the university's learning management system (LMS), combined with student submission scores and demographic information, was used to identify early warning signs of dropouts and differentiate between successful and unsuccessful students. Students were grouped into three categories: i) non-completers (less than 50% at the final grade), ii) low completers ($<75 \%$), and iii) high completers ($>75 \%$). Five machine learning models were trained to identify the most important features contributing to course success. Our work illustrates that non-completers show distinct behavior from the start, particularly in lab performance, where they consistently underperformed compared to other groups. While dropout predictions can be made with an accuracy of 80% as early as week 2, later semester activities and assessments have a greater impact on final grades. Engagement in lab units proved to be crucial for student success, with non-completers exhibiting performance differences early on. These results underscore the importance of monitoring lab engagement to identify students at risk of dropping out and suggest that early intervention within the course framework could help mitigate attrition rates.