An Interpretable Machine Learning Framework for High-Potential Student Identification in Vocational Education
Year: 2026
Type: conference-paper
Abstract: Identifying high-potential students in vocational education is important for targeted talent cultivation and efficient allocation of educational resources. However, conventional evaluation methods often rely on manual judgment or single-score criteria, making it difficult to capture students’ multidimensional abilities under imbalanced data conditions. To address this issue, this paper proposes an interpretable machine learning framework for high-potential student identification in vocational education. Using students’ academic performance records as input features and competition award status as an observable proxy for student potential, the study formulates the task as a binary classification problem. Four models, including balanced Logistic Regression, balanced Random Forest, XGBoost, and class-weighted XGBoost, are compared on a real-world dataset of 165 students with 155 numerical features. Experimental results from a stratified train-test split suggest that balanced Logistic Regression achieved the most competitive overall performance on this dataset, with an accuracy of 0.9697, precision of 1.0000, recall of 0.5000, F1-score of 0.6667, and AUC of 0.9677. Feature importance analysis further indicates that several professional and practice-oriented courses may contribute substantially to high-potential student identification. Due to the small sample size and the limited number of positive cases, these findings should be interpreted as preliminary. Nevertheless, the results suggest that interpretable machine learning may provide useful decision support for early identification and targeted cultivation of promising students in vocational education.
Language: English
Subfield: Computer Science Applications
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7213132515
