基于症状等多源数据的急性呼吸道传染病发病趋势预测模型的构建与评价  

Construction and evaluation of a prediction model for the trend of acute respiratory infectious diseases based on multi-source data including symptom surveillance

作　　者：邬安琪 文泽轩 吴强松 汪晨夕 施建华 WU An-qi;WEN Ze-xuan;WU Qiang-song;WANG Chen-xi;SHI Jian-hua(Xuhui Center for Disease Control and Prevention,Shanghai 200237,China;不详)

机构地区：[1]徐汇区疾病预防控制中心,上海200237 [2]复旦大学上海市重大传染病和生物安全研究院 [3]复旦大学公共卫生学院流行病学教研室/公共卫生安全教育部重点实验室

出　　处：《现代预防医学》2025年第2期220-226,共7页Modern Preventive Medicine

基　　金：2021年度徐汇区医学科研项目(SHXH202147)。

摘　　要：目的构建基于多源数据的遗传算法优化的支持向量机(Genetic Algorithm optimized Support Vector Machine,GA-SVM)模型预测急性呼吸道传染病并评价其预测效果,为建立呼吸道传染病早期预警体系提供参考。方法根据2020—2022年上海市徐汇区的症状监测、气象及大气污染和严格指数数据,在潜在预测变量中挑选最佳延迟周数的预测变量后,筛选出预测重要性最强的变量作为自变量。按1∶4比例将全时间序列划分为验证集和训练集,利用遗传算法优化参数,以呼吸道传染病每周新增病例数为因变量构建GA-SVM模型。采用均方根误差、平均绝对百分比误差、预测相关系数和决定系数对模型预测结果进行评价。结果预测重要性最强的变量为:延迟2周的严格性指数、延迟1周的症状监测病例数、延迟1周的最高气温、延迟2周的学校活动和延迟1周的臭氧(O3)指数。建立的GA-SVM模型最优参数C=18.04,γ=0.1754,模型的平均均方根误差为6.362,平均绝对百分比误差为24.59%,平均预测相关系数和决定系数分别为0.896和0.804。结论该模型对于徐汇区急性呼吸道传染病报告病例数有着良好的预测效果,证实了GA-SVM应用于基于症状等多源数据实现呼吸道传染病预测的可行性,为多源数据应用于传染病早期预警提供方法参考。Objective Toconstruct a Genetic Algorithm optimized Support Vector Machine(GA-SVM)model based on multi-source data predicting acute respiratory infectious diseases and toevaluate its predictive effectiveness,providing a reference for establishing an early warning system for respiratory infectious diseases.Methods Symptom surveillance cases,meteorological and atmospheric pollution,data and stringency index obtained from 2020 to 2022 were used as modeling and forecasting samples,respectively.By picking up the optimum lagging week number of the potential predictive variables and filter out the most important variables successively,the independent variables were obtained.Then the full time series data were divided into validation set and training set in a 1:4 ratio.The parameters were optimized by genetic algorithm.We used the weekly number of new cases of respiratory infectious diseases as the dependent variable to structure the GA-SVM model.The performance was evaluated based on the following metrics:root mean square error(RMSE),meansabsolute percentage error(MAPE),predictive correlation coefficient(PCC)and R-squared(R2).Results The most important variables were stringency index with 2-weeks-lag,symptom surveillance cases with 1-week-lag,maximum temperature with 1-week-lag,school activities with 2-weeks-lag and O3 index with 1-week-lag.The GA-SVM model performed best when C=18.04,γ=0.1754 while average RMSE=6.362,average MAPE=24.59%,average PCC=0.896 and average R2=0.804.Conclusions The model showsgood predictive performance for the reported cases of acute respiratory infectious diseases in Xuhui District,which confirms the feasibility of applying GA-SVM to multi-source data based on symptom monitoring for predicting respiratory infectious diseases,providing methodological references for the application of multi-source data in the early warning of infectious diseases.

关 键 词：多源数据 支持向量机 急性呼吸道传染病 症状监测 预测 

分 类 号：R181.3[医药卫生—流行病学]