Predictive modeling of student success

The creation of machine learning-based predictive models of student success has risen as a central activity in the fields of artificial intelligence in education, learning analytics, and educational data mining. Fueled by the high availability of data on learner interactions with technology, as well as the broad availability of computational methods to deal with these data, predictive modeling in education tends to have a lesser emphasis on producing learning theories and causal interpretations of data and a greater emphasis on predicting accurate coarse-grained learning outcomes for individual students. This chapter presents an overview of the common data sources, discusses the recent developments, and identifies the critical issues in predictive modeling of student success. Open challenges in the area include model explainability, fairness and bias in models, model generalizability and transfer to new circumstances, and the educational actionability of models.