多源数据驱动的细粒度传染病预测模型

A Fine-Grained Infectious Disease Prediction Model Driven by Multi-Source Data

目的 研究基于多源数据的传染病细粒度预测模型,为传染病精准防控提供依据。方法 基于传染病历史确诊数据以及来自医疗机构和社会的外部数据,构建多源多粒度时空网络(MMSTNet)。MMSTNet充分融合了不同空间粒度的数据,采用图注意力网络捕捉空间相关性,采用门控循环单元捕捉时间相关性,预测未来细粒度传染病确诊人数。结果 MMSTNet在各预测天数下预测误差均小于基线模型,其平均绝对误差比最佳基线模型误差降低14.4%。结论 融合来自医疗机构和社会的外部数据、考虑区域间的空间相关性,能够有效提升细粒度传染病预测准确性。

Objective To develop a fine-grained infectious disease prediction model based on multi-source data,providing a basis for precise prevention and control of infectious diseases.Methods Based on historical confirmed case data of infectious diseases and external data from medical institutions and society,we propose a Multi-source Multi-grained Spatio-temporal Network(MMSTNet).It fully integrates data of different spatial granularity,leverages graph attention networks to capture spatial correlations,and gated recurrent units to capture temporal correlations,and predicts the number of fine-grained confirmed cases of infectious diseases in the future.Results The prediction error of MMSTNet is smaller than all baselines over all prediction days,with its mean absolute error reduced by 14.4%compared to the best baseline.Conclusion Integrating external data from medical institutions and society,and considering spatial correlations between regions,can effectively improve the accuracy of fine-grained infectious disease predictions.