Academic failure is one of the most important problems in higher education institutions around the world. In a continuous cycle, universities admit new students and graduate others, but the road to success in academia is teeming with failures. This work envisions a predictive model to identify students at risk of failure in time to reverse the situation.

A tremendous number of papers that apply machine learning (ML) have already been published, reporting progress in practical developments. ML applications in the educational field are no different. The paper presents a comprehensive survey of studies on predicting student outcomes, and situates the work in this wider research context. The paper proposes a support vector machine (SVM) prediction model that leverages peer instruction data to make predictions for multiple courses across multiple institutions.

The authors performed experiments on data from three instructors at two public universities in North America. Evaluations consider the standard receiver operating characteristic (ROC) metric. On average, the ROC curve achieved 0.70, which means there is a 70 percent chance that the model will be able to distinguish between students at risk or not.

While someone can criticize the not so high accuracy of the model, as highlighted in the paper, prior predictive models are based on data that is difficult to get and applied to a single course/institution. This paper advances the state of the art, proposing a model to deal with multiple real-world settings.