New Validation Protocol Assesses Knowledge Tracing Model Explanations

VERA Brief

AI-generated. Grounded in the article and its cited sources.

A new validation protocol has been proposed for knowledge tracing models to enhance the transparency of their student performance predictions. The protocol evaluates predictive competitiveness, explanation stability, and retraining-based faithfulness.

Key facts

• A new validation protocol for knowledge tracing models has been proposed in a study on arXiv.
• The protocol assesses predictive competitiveness, explanation stability, and retraining-based faithfulness.
• Thirteen behavioral features from five pedagogical themes were used, derived from the ASSISTments 2009 and 2012 datasets.
• Data leakage issues in the ASSISTments 2009 dataset were addressed by rebuilding it, which affected model AUC and explanation order.
• An Extreme Gradient Boosting model explained using TreeSHAP was compared to four deep learning baselines.