Student Performance Prediction (SPP) models and tools are useful for quickly identifying at-risk students in online courses and enable the provision of personalized learning plans and assistance. Additionally, they give educators and course managers the information they need to recognize the programs that require improvement. High accuracy is essential for such tools, but understanding the reasons of their predictions is equally important to ensure fairness and build trust in their adoption. Although many SPP models and tools have been proposed so far by different researchers, very few of them take explainability into account. This research proposes an SPP approach that is both effective and explainable. Based on demographic, administrative, engagement, and intra-course outcome data, it enables the prediction of student performance in terms of success/failure and final grade. It supports multiple machine learning models and includes post-hoc techniques for explainability capable of justifying the behavior of the whole system as well as its individual predictions.
Explainable Prediction of Student Performance in Online Courses
Capuano N.;
2023-01-01
Abstract
Student Performance Prediction (SPP) models and tools are useful for quickly identifying at-risk students in online courses and enable the provision of personalized learning plans and assistance. Additionally, they give educators and course managers the information they need to recognize the programs that require improvement. High accuracy is essential for such tools, but understanding the reasons of their predictions is equally important to ensure fairness and build trust in their adoption. Although many SPP models and tools have been proposed so far by different researchers, very few of them take explainability into account. This research proposes an SPP approach that is both effective and explainable. Based on demographic, administrative, engagement, and intra-course outcome data, it enables the prediction of student performance in terms of success/failure and final grade. It supports multiple machine learning models and includes post-hoc techniques for explainability capable of justifying the behavior of the whole system as well as its individual predictions.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.