Data-driven higher-education analytics

Higher education in the Gulf Cooperation Council region faces student underachievement despite large budgets [5], and the literature on retention, progression and graduation in health professions education is scarce [3].

When universities moved online during the COVID-19 pandemic, often without preparation, questions arose about student satisfaction [4] and about how instructors could anticipate learner performance before the final assessment [2]. A further question is whether performance in one course of a linked chain predicts performance in the next [1].

Final grades were analysed for chains of linked courses with high D, F and W rates in Arabic (n = 11,780), English (n = 7,714) and mathematics (n = 1,367) from 2012 to 2016 [1].

Longitudinal data for students enrolled in the five colleges of Qatar University’s Health Cluster between 2015 and 2021 were analysed with descriptive statistics and an XGBoost predictive model with feature-importance scores [3].

A two-pathway deep learning model classifies learner performance from online clickstreams, transformed into images with the Gramian Angular Field, together with demographic and assessment data, and was evaluated on the Open University Learning Analytics Dataset [2].

A survey of 2,354 students modelled satisfaction with online learning from seven factors and four demographic variables [4], and Academic Success Plan responses from 5,040 at-risk students identified the causes of underachievement [5].

Evidence from large institutional datasets on the drivers of student performance, retention, satisfaction and underachievement [1, 3, 4, 5], and predictive models, gradient boosting [3] and deep learning [2], for identifying students who may need support.

Between 60% and 75% of low-performing students in one link of a course chain also performed poorly in the next link [1].

First-year performance, followed by high-school GPA, ranked as the most important predictors of retention, progression and graduation in health majors [3].

All seven factors except assessment significantly predicted satisfaction with online learning; among the demographic variables only GPA did, and the regression model had a coefficient of determination of 0.723 [4].

Students reported academic factors as the most frequent cause of underachievement and social-adjustment factors as the least frequent [5]; including demographic and assessment data improved the prediction of learner performance [2].

Academic advising, student support, engagement and communication [3], proactive measures to assist learners before the final assessment [2], and intervention programmes for students at risk [5].