A Game Theoretic Framework for Interpretable Student Performance Model

Abstract

Machine learning is used in many contexts these days. And they’ve been integrated into the decision-making process in many critical areas, some applications including predicting at-risk students and automating student enrollment. From these applications, it is clear that machine learning models have a major impact on students’ professional success, therefore, it is imperative that the student performance model is well understood and free of any bias and discrimination. The kinds of decisions and predictions made by these machine-learning-enabled systems become much more profound and, in many cases, critical to students’ professional success. Various higher education institutions rely on machine learning to drive their strategy and improve their students academic success. Therefore, the need to trust these machine learning-based systems is paramount, and building a model that educational decision-makers who may not be familiar with machine learning can understand is critical. But sometimes, even for the experts in machine learning, it becomes difficult to explain certain predictions of the so-called “black box models”. Therefore, there is a growing need for easy interpretation of a complex black box model. Therefore, this study aims to provide a framework for an interpretable student performance model by introducing a local model-agnostic interpretability method shap value, which is a novel explanatory technique that explains the predictions of any classifier in an interpretable and faithful way by opening the black -box model and explaining how the final result came out and which parts of the model are responsible for certain predictions. By understanding how the student performance model works, education decision-makers can have a greater advantage and be smarter about what they should be doing.