Academic performance warning system based on data driven for higher education.

Area of Science:

• Educational Technology
• Data Science
• Machine Learning in Education

Background:

• Academic probation poses significant challenges for university students.
• Existing methods for identifying at-risk students often lack predictive accuracy.
• There is a need for proactive interventions to mitigate academic probation rates.

Purpose of the Study:

• To develop and evaluate a machine learning-based academic warning system.
• To leverage student academic performance data for early risk detection.
• To create a scalable and adaptable dataset for predicting academic warning status.

Main Methods:

• Utilized massive educational data sources and machine learning techniques.
• Developed a flexible and scalable dataset with detailed calculation formulas.
• Employed feature generation, selection, data balancing, and model selection strategies.
• Implemented a two-stage warning system using Support Vector Machine and LightGBM algorithms.

Main Results:

• Achieved an F2-score of over 74% for early semester warnings using Support Vector Machine.
• Achieved an F2-score of over 92% for pre-final examination warnings using LightGBM.
• The developed dataset is reusable and reconstructible for other institutions.

Conclusions:

• The proposed two-stage academic warning system effectively predicts students at risk of probation.
• Machine learning models, particularly LightGBM, demonstrate high efficacy in identifying at-risk students.
• The system provides a valuable tool for universities to implement timely interventions and improve student outcomes.