Introduction
While Artificial Intelligence (AI) has seen rapid adoption in higher education and corporate sectors, its integration into K-12 environments remains limited. The primary barriers include significant data heterogeneity across different school systems, stringent privacy requirements for minor-related data, and a lack of unified platforms that address the diverse needs of educators. DMP_AI is a comprehensive, multi-module platform designed to bridge these gaps by providing predictive analytics, behavioral monitoring, and privacy-preserving recommendations.
System Architecture and Methodology
The DMP_AI system is built upon a "Data Lake" architecture that facilitates the ingestion of multi-source data, including grades, attendance, and disciplinary records. To ensure compliance with privacy standards, the system incorporates a dedicated de-identification and encryption module.
The platform's core functionality is divided into several key technical components:
- Predictive Analytics: Standard machine learning classifiers, such as Random Forest and SVR, are employed to forecast student performance in both internal assessments and public examinations (e.g., the DSE).
- NLP for Individualized Education Plans (IEPs): Natural Language Processing techniques, including Part-of-Speech tagging and keyword extraction, allow counselors to analyze textual data within IEPs, identifying common trends in special education needs.
- Heterogeneity-aware Federated Recommendation (HFRec): To recommend elective courses across schools without centralizing sensitive data, the system utilizes a federated learning scheme. This approach models student-course interactions using heterogeneous graphs while keeping raw data local to each school.
The Early Warning System (EWS)
A critical component of DMP_AI is its Early Warning System, which translates complex predictive data into intuitive visualizations for teachers. Using a "traffic light" metaphor (Red/Yellow/Green), the system alerts educators to potential declines in academic performance or behavioral issues.
Pilot Study and Key Insights
The system was evaluated through a pilot study involving eight schools in Hong Kong. Data was sourced from the Web/Cloud-SAMS (School Administration and Management System), reflecting the complex, real-world data environments of diverse schools.
Key findings from the study include:
- High User Satisfaction: The system received an average rating above 3.0/5.0, with the user interface (UI) being the most highly-rated feature (3.86/5).
- The Explainability Gap: While teachers found the system helpful, their self-reported understanding of the underlying AI was relatively low (2.89/5). This highlights a critical need for improved model interpretability and AI literacy training for educators.
- Privacy-Preserving Collaboration: Federated learning successfully addressed the "cold start" problem for smaller schools by allowing them to benefit from global knowledge without compromising student privacy.
Conclusion and Future Work
DMP_AI demonstrates the feasibility of deploying sophisticated AI tools in the K-12 sector. Future development will focus on enhancing AI literacy among teachers, improving data consistency across institutions, and expanding the platform to mobile devices to better support active classroom management.
Summary updated