Supervised diagnostic classification of cognitive attributes using data augmentation.

Area of Science:

• Artificial Intelligence
• Educational Measurement
• Machine Learning

Background:

• Machine learning (ML) offers data-driven decision-making across sectors, including educational assessment.
• Diagnostic Classification Models (DCMs) diagnose student cognitive states but face challenges with latent, unlabeled data.
• Existing ML applications in DCMs often rely on small, expert-labeled datasets.

Purpose of the Study:

• To propose a supervised diagnostic classification model with data augmentation (SDCM-DA).
• To enhance the classification accuracy of students' cognitive states in educational assessments.
• To address the challenge of limited labeled data in applying ML to DCMs.

Main Methods:

• Developed a supervised diagnostic classification model with data augmentation (SDCM-DA).
• Constructed a data generation model using probabilities of correct responses from expert-labeled data.
• Conducted a simulation study comparing SDCM-DA with traditional methods using only expert-labeled data.

Main Results:

• Data augmentation substantially improved classification accuracy.
• Enhanced performance was observed even with small, error-prone labeled samples.
• Effective student classification was achieved without needing an explicit underlying response model.

Conclusions:

• SDCM-DA effectively leverages augmented data to improve diagnostic classification.
• The method is robust to limited and imperfect expert labels and lower-quality test items.
• This approach advances ML applications in educational assessment by overcoming data limitations.