A Chinese Dataset for Training AI Models to Assess Procedural Accuracy and Scientific Compliance in VirtualMedical Simulation

Abstract Virtual medical simulation has emerged as an indispensable pedagogical modality in health professions education, particularly in resource-constrained settings where clinical access remains limited and standardized patient resources are scarce. However, the challenge of ensuring procedural accuracy and scientific compliance during simulation-based learning through automated assessment remains inadequately addressed. This paper presents the Virtual Medical Simulation Training (VMST) dataset, a comprehensive Chinese-language corpus specifically designed to train artificial intelligence models for identifying and evaluating procedural accuracy and standardization in virtual medical simulation environments. The VMST dataset comprises 15,847 annotated operation sequences collected from 2,156 vocational medical students across 12 simulation training modules, encompassing nursing procedures, clinical examination techniques, and emergency response protocols. Each operation sequence includes timestamped action logs, expert-annotated accuracy labels, compliance scores based on standardized protocols, and detailed error classifications. Technical validation demonstrates that deep learning models trained on VMST achieve 89.3% accuracy in identifying procedural errors and 91.7% precision in compliance assessment, substantially outperforming baseline methods. This dataset addresses a critical gap in AI-assisted medical education assessment and establishes a foundation for developing intelligent tutoring systems capable of providing real-time feedback to students in virtual simulation environments.