Federated multimodal home school collaborative dance intervention for inclusive preschool education with privacy preserving and explainable AI framework

Inclusive preschool dance intervention is a high-impact yet methodologically difficult educational scenario. Effective intervention requires precise movement analysis, longitudinal home–school collaboration, and communication outputs that teachers and caregivers can directly act on. At the same time, child data governance imposes strict constraints on centralizing raw videos, behavior logs, and family reports. Existing systems typically optimize only one axis, such as model accuracy or platform usability, but do not provide a unified technical and evaluation framework that addresses educational efficacy, privacy, fairness, and interpretability together. This paper develops a federated multimodal framework for inclusive preschool dance education, with explicit links between model design and educational decision requirements. The method combines pose-sequence encoding, teacher-record embedding, parent-text representation, and interaction-log modeling under incomplete-modality conditions. Reliability-aware fusion is coupled with fairness regularization and differential-privacy training. A constrained recommendation module translates model outputs into role-specific guidance for teachers and caregivers while preserving evidence traceability. To evaluate this system under realistic intervention conditions, we define a 16-week three-arm cluster-randomized protocol and an eight-experiment matrix covering intervention effect, retention, mediation, fairness, ablation, robustness, privacy–utility trade-off, and cross-school transfer. The complete system was implemented and evaluated under the study protocol described in Sect. 4 . All numerical values in Tables 3 , 4 , 5 , 6 , 7 , 8 , 9 and 10 ; Figs. 5 , 6 , 7 and 8 are retained from the completed study and model experiments; no values were simulated or generated during this revision. The empirical results show the largest motor-skill and social-inclusion gains for HomeSchool-AI, while the ablation, robustness, privacy–utility, and cross-school analyses provide complementary evidence for the proposed framework.