The application of machine learning in predicting student performance in university engineering programs: a rapid review

Abstract

Background: In recent years, the application of machine learning (ML) to predict student performance in engineering education has expanded significantly, yet questions remain about the consistency, reliability, and generalisability of these predictive models. Objective: This rapid review aimed to systematically examine peer-reviewed studies published between January 1, 2019, and December 31, 2024, that applied machine learning (ML), artificial intelligence (AI), or deep learning (DL) methods to predict or improve academic outcomes in university engineering programs. Methods: We searched IEEE Xplore, SpringerLink, and PubMed, identifying an initial pool of 2,933 records. After screening for eligibility based on pre-defined inclusion criteria, we selected 27 peer-reviewed studies for narrative synthesis and assessed their methodological quality using the PROBAST framework. Results: All 27 studies involved undergraduate engineering students and demonstrated the capability of diverse ML techniques to enhance various academic outcomes. Notably, one study found that a reinforcement learning-based intelligent tutoring system significantly improved learning efficiency in digital logic courses. Another study using AI-based real-time behavior analysis increased students’ exam scores by approximately 8.44 percentage points. An optimised support vector machine (SVM) model accurately predicted engineering students’ employability with 87.8% accuracy, outperforming traditional predictive approaches. Additionally, a longitudinally validated SVM model effectively identified at-risk students, achieving 83.9% accuracy on hold-out cohorts. Bayesian regression methods also improved early-term course grade prediction by 27% over baseline predictors. However, most studies relied on single-institution samples and lacked rigorous external validation, limiting the generalisability of their findings. Conclusion: The evidence confirms that ML methods—particularly reinforcement learning, deep learning, and optimised predictive algorithms—can substantially improve student performance and academic outcomes in engineering education. However, methodological shortcomings related to participant selection bias, sample sizes, validation practices, and transparency in reporting require further attention. Future research should prioritise multi-institutional studies, robust validation techniques, and enhanced methodological transparency to fully leverage ML’s potential in engineering education.