A Multimodal Temporal Diagnostic and Personalized Intervention Framework for Sustainable Teacher Preparation: Evidence from TPACK Development in Pre-Service Kindergarten Teachers

Artificial intelligence is reshaping teacher education, yet many existing approaches still rely on static assessment and provide limited support for interpretable and actionable diagnosis. This limitation is especially important for pre-service kindergarten teachers, whose TPACK development is dynamic, multidimensional, and context-sensitive. To address this gap, this study proposes a multimodal temporal diagnostic and personalized intervention framework for sustainable teacher preparation. Using longitudinal data from 186 pre-service kindergarten teachers across four developmental stages over three academic years, the framework integrates self-report, behavioral, performance, and reflective evidence to model developmental trajectories, identify latent states, and generate explainable support recommendations. The results show that overall TPACK increased from 3.21 to 3.73, with the largest gains in technology-related integration dimensions. Compared with representative baselines, the proposed framework achieved the best diagnostic performance (MAE = 0.171, RMSE = 0.238, Accuracy = 0.821, Macro-F1 = 0.806). These findings suggest that AI can support teacher education not only through more accurate diagnosis, but also through interpretable and development-sensitive intervention.