A Medical Data Generative Model based on Knowledge Graph Attention Network

With the rapid growth of medical data, the challenges between enhancing the quality of medical data and ensuring patient privacy are becoming increasingly prominent. Conventional data anonymization methods often fail to effectively protect medical data, prompting us to seek advanced techniques like differential privacy and data synthesis. This study introduces the Knowledge Graph Attention Network (KGAT) to generate synthetic medical data, aiming to enhance the interpretability of the model. The methods include classifying data fields, preprocessing the data, and using KGAT and Decision Tree to generate core fields. The synthetic medical records are evaluated through JS divergence and Wasserstein distance metrics. Preliminary results indicate that the generation performance of KGAT on most core fields is close to the original data. This will provide a new direction for the field of medical data generation.