Data-Driven Multimodal Assessment Model for Mathematical Writing Proficiency

This research aims to construct and confirm a multimodal data-driven assessment framework of mathematical writing proficiency through the conglomeration of English linguistic capabilities and mathematical symbolic analysis. A total of 240 Grade 910 students were sampled and 720 written responses made. The spaCy v3.7, scikit-learn v1.4, SymPy v1.12 and R lavaan v0.6-17 were used to perform feature extraction and modelling. Techniques were NLP-based feature extraction, symbolic validation, Confirmatory Factor Analysis and the regression model. Findings yielded three latent dimensions namely Conceptual Clarity, Logical Coherence, and Symbolic Accuracy that had a good model fit (CFI ≈ 0.94). The predictive models were very accurate (R 2 0.87) and far much better than rubric-only predictive models. The most effective predictor was the symbolic accuracy. The results reveal that the combination of linguistic and mathematical characteristics enhances reliability and validity when it comes to testing the proficiency of mathematical writing.