Construction and Empirical Research on Classroom Teaching Evaluation Model of Vocational Education Based on Artificial Intelligence Multimodal Data Fusion

Against the backdrop of policies promoting quality enhancement and excellence in vocational education alongside educational evaluation reforms, traditional classroom teaching assessments struggle to comprehensively reflect teaching quality due to limitations such as single-dimensional data and insufficient technological empowerment. To address this, this study constructs an AI-driven multimodal data fusion model for classroom teaching evaluation. This model encompasses a five-layer technical architecture—data collection, preprocessing, feature extraction, metric calculation, and result feedback—integrating three types of multi-source heterogeneous information: teaching behavior data (based on DBSCAN clustering analysis), physiological state data (combined with ResNet-18 emotion recognition), and dynamic effect data (using TransE knowledge transfer evaluation). This enables comprehensive and precise diagnosis of the classroom teaching process. Empirical validation was conducted in smart classroom environments using S-T behavioral analysis, engagement curves, and the Rt-Ch model. Results demonstrate the model’s effectiveness in identifying critical issues such as imbalanced teaching pacing and fluctuating student engagement, significantly enhancing the comprehensiveness and objectivity of evaluations. This research not only provides a replicable technical pathway for diagnosing vocational education classroom quality but also offers theoretical support and practical examples for the deep integration of artificial intelligence and educational assessment.