A Critical Examination of Machine Learning as a Tool to Predict Performance of Students in CS1

This full research paper presents a systematic literature review of research using machine learning techniques to predict student performance in introductory programming courses. The overarching research question is: How does empirical research using machine learning approach the prediction of student performance in introductory computer science courses (CS1)? The focus is on how knowledge from educational science is incorporated alongside with ethical and gender considerations. Only peer-reviewed articles, published in journals or conference proceedings between 2017 and mid 2020, reporting on empirical studies that used data on more than 30 students are included. This study addresses prevalent shortcomings in empirical CS education research, noting often inadequate descriptions of data selection, processing, and the representation and diversity of sample sizes that can limit the utility of results. It underscores the frequent omission of ethical considerations regarding students' data consent and the potential negative impacts on students' educational trajectories. Additionally, many studies fail to incorporate the educational context or address gender-related issues adequately, disconnecting the models from established knowledge about women in computer science.