Machine learning for university management
Micro Cluster Learning to predict "active" students
Bibliographic Data
| ID | 20349136 |
|---|---|
| Authors | Alexander Karl Ferdinand Loder (0009-0008-8906-8410, University of Graz, corresponding author) |
| Year | 2025 |
| Volume | 85 |
| Pages | 101463 |
| Publication date | 2025-06-01 |
| Peer Reviewed | Yes |
| Open Access | Yes |
| Type | ARTICLE |
| Venue | Studies In Educational Evaluation (JOURNAL) |
| Journal identifiers | ISSN: 0191-491X • E-ISSN: 1879-2529 |
| Publisher | Elsevier BV (PUBLISHER) |
| DOI | 10.1016/j.stueduc.2025.101463 |
| OpenAlex | W4409573211 |
| Language | EN |
| References cited | 61 |
The strategy process of universities can span several years into the future with prediction of student performance being an important aspect for university governance. “Micro Cluster Learning” is proposed, by applying a hybrid of machine learning and ARIMA models to micro clusters of a university’s administrative data. The aim was to predict a performance indicator one to three academic years in the future and to compare the results of five years and to official statistics. Micro clusters were generated and a stack of 20 machine learning algorithms was applied to each cluster. The algorithms and their hyperparameter settings were determined in an explorative manual pre-selection process. The results show deviations from the official statistics between 2 % and 8 % ( SD = 6 %) for one academic year in the future, 1–29 % ( SD = 19 %) for two and 1–17 % ( SD = 11 %) for three years. Model performance correlated with increasing details in the micro clusters and was better in larger micro clusters. The method is very flexible and can be used in a multitude of different university settings worldwide and for different outcomes of interest, e.g., grades or student status. However, the flexibility goes along with a tedious setup and very long runtimes. Future improvements with increased automation are warranted and “meta-procedures” should be developed that can perform automated resampling and hyperparameter tuning on a stack of algorithms. The method presented in this study contributes to preventive university management in different countries and university systems. • A hybrid machine learning and forecasting approach is proposed for university student performance prediction. • “Micro Cluster Learning” algorithm is proposed. • The model is among the largest and most complex models for student performance prediction to date. • The performance outcomes are predicted one, two and three years in the future and results are compared to official statistics. • This method can be adapted to university systems worldwide
Cluster grouping · Machine learning · Mathematics education · Anomaly Detection Techniques and Applications · Computer Science · Data Stream Mining Techniques · Online Learning and Analytics · Psychology
Interrupted time series analysis using autoregressive integrated moving average (Arima) models
Learning representations by back-propagating errors
Student Drop‐out from G erman Higher Education Institutions
Labor demand in the time of Covid-19
The Elements of Statistical Learning
Multivariate Adaptive Regression Splines
Student dropout from universities in Europe
A new family of power transformations to improve normality or symmetry
Building Predictive Models in R Using the caret Package
Regularization Paths for Generalized Linear Models via Coordinate Descent
Random Forests
Regression Models and Life-Tables
A Systematic Literature Review of Student’ Performance Prediction Using Machine Learning Techniques
An Intelligent Nudging System to Guide Online Learners
Evaluating Impacts of Performance Funding Policies on Student Outcomes in Higher Education
It was nothing to do with the university, it was just the people
| Citation velocity | historical |
|---|---|
| Highly cited | No |