Tri-Branch Convolutional Neural Networks for Top-k Focused Academic Performance Prediction

Area of Science:

• Educational Data Mining
• Machine Learning in Education
• Artificial Intelligence in Education

Background:

• Academic performance prediction is crucial for personalized teaching and early warnings.
• Existing methods often focus on overall prediction accuracy, neglecting the identification of at-risk students.
• Student behavior patterns, captured through campus smartcard records, offer valuable insights.

Purpose of the Study:

• To predict academic performance by analyzing student behavior trajectories from smartcard data.
• To develop a novel deep learning model capable of capturing complex behavioral patterns.
• To improve the accuracy of identifying academically at-risk students through a top-k ranking approach.

Main Methods:

• Utilized campus smartcard records to mine and analyze student behavior trajectories.
• Designed a tri-branch convolutional neural network (CNN) with specialized convolutions and attention mechanisms.
• Implemented a top-k focused loss function to prioritize the accurate ranking of at-risk students.

Main Results:

• The proposed tri-branch CNN effectively captures persistence, regularity, and temporal distribution of student behavior.
• The top-k focused loss significantly enhances the accuracy in identifying academically at-risk students.
• The approach demonstrates substantial performance improvements over existing methods on a large-scale dataset.

Conclusions:

• Mining student behavior trajectories from smartcard data is effective for academic performance prediction.
• The novel tri-branch CNN architecture and top-k focused loss offer a superior approach for identifying at-risk students.
• This research provides a valuable tool for early intervention and personalized educational strategies.