Early Prediction of Student Dropout in Higher Education Using Machine Learning Models
Saved in:
| Title: | Early Prediction of Student Dropout in Higher Education Using Machine Learning Models |
|---|---|
| Language: | English |
| Authors: | Or Goren, Liron Cohen, Amir Rubinstein |
| Source: | International Educational Data Mining Society. 2024. |
| Availability: | International Educational Data Mining Society. e-mail: admin@educationaldatamining.org; Web site: https://educationaldatamining.org/conferences/ |
| Peer Reviewed: | Y |
| Page Count: | 11 |
| Publication Date: | 2024 |
| Document Type: | Speeches/Meeting Papers Reports - Research |
| Education Level: | Higher Education Postsecondary Education |
| Descriptors: | Dropout Characteristics, Prediction, Potential Dropouts, Student Characteristics, At Risk Students, College Students, Artificial Intelligence, Predictor Variables, Intellectual Disciplines, Learning Management Systems, Student Behavior, Grades (Scholastic) |
| Abstract: | The problem of student dropout in higher education has gained significant attention within the Educational Data Mining research community over the years. Since student dropout is a major concern for the education community and policymakers, many research studies aim to evaluate and uncover profiles of students at-risk of dropping out, allowing timely intervention. Many students who eventually drop out of university display signs during their first year of studies, thus early identification of these students is both beneficial and feasible. Prediction studies typically use administrative data, encompassing a wide range of student attributes, as well as learning behavior data from an institute's learning management system. In this study, we evaluate the early prediction of at-risk students within a traditional classroom setting at a higher education institute through machine learning methods, particularly Neural Networks and XGBoost. In an attempt to generalize our results, we compare different fields of study, namely exact and social sciences. Our results show that integrating learning behavior has little effect on prediction quality metrics. We also propose a novel aggregated 'studentship' feature, containing both cognitive and social elements, which improves the AUC metric for XGBoost at early stages of the semester. Furthermore, our results indicate that prediction patterns vary among different fields of study. We believe this research presents a compelling case study and provides several valuable insights into the expanding corpus of studies on dropout prediction in higher education. [For the complete proceedings, see ED675485.] |
| Abstractor: | As Provided |
| Entry Date: | 2025 |
| Accession Number: | ED675569 |
| Database: | ERIC |
| Abstract: | The problem of student dropout in higher education has gained significant attention within the Educational Data Mining research community over the years. Since student dropout is a major concern for the education community and policymakers, many research studies aim to evaluate and uncover profiles of students at-risk of dropping out, allowing timely intervention. Many students who eventually drop out of university display signs during their first year of studies, thus early identification of these students is both beneficial and feasible. Prediction studies typically use administrative data, encompassing a wide range of student attributes, as well as learning behavior data from an institute's learning management system. In this study, we evaluate the early prediction of at-risk students within a traditional classroom setting at a higher education institute through machine learning methods, particularly Neural Networks and XGBoost. In an attempt to generalize our results, we compare different fields of study, namely exact and social sciences. Our results show that integrating learning behavior has little effect on prediction quality metrics. We also propose a novel aggregated 'studentship' feature, containing both cognitive and social elements, which improves the AUC metric for XGBoost at early stages of the semester. Furthermore, our results indicate that prediction patterns vary among different fields of study. We believe this research presents a compelling case study and provides several valuable insights into the expanding corpus of studies on dropout prediction in higher education. [For the complete proceedings, see ED675485.] |
|---|
