Self-paced graph memory for learner GPA prediction and it's application in learner multiple evaluation | JoVE Visualize

Area of Science:

Educational Technology
Data Science in Education
Machine Learning Applications

Background:

Current teaching evaluation relies heavily on Grade Point Average (GPA), which can obscure differences in learning abilities and patterns among students.
The limitations of GPA-based assessment hinder the development of truly personalized learning experiences.
Identifying distinct learning patterns is crucial for effective educational strategies.

Purpose of the Study:

To propose a data-driven assessment strategy that supplements traditional GPA evaluation for more nuanced understanding of student learning.
To develop a novel learning performance prediction model integrating self-paced learning and graph memory neural networks.
To identify students with unique learning patterns that may be masked by uniform GPA scores.

Main Methods:

Developed a 'self-paced graph memory network' by integrating self-paced learning and graph memory neural networks for learning performance prediction.
Employed a t-test approach, inspired by outlier detection in linear regression, to identify students with significantly different loss values.
Analyzed learning process data to differentiate learning patterns within the same GPA level.

Main Results:

The proposed data-driven strategy effectively identifies students with distinct inherent learning patterns, irrespective of their GPA.
Analysis revealed that students identified with unique learning patterns exhibit a distribution across various GPA levels.
The method successfully addresses the shortcomings of GPA-only evaluation models by providing deeper insights into learning behaviors.

Conclusions:

The data-driven assessment strategy offers a more rational and effective supplement to traditional GPA evaluation for personalized learning.
The developed self-paced graph memory network and t-test approach provide a robust framework for analyzing student learning data.
Validation through protein classification and student performance prediction experiments confirms the method's rationality and effectiveness in student data modeling.