多层感知器神经网络在手足口病预测模型中的应用

Application of multi-layer perceptron neural network in the prediction model of hand-foot-mouth disease

目的研究多层感知器神经网络手足口病预测的应用,为手足口病预测提供依据。方法选取浙江省2008年5月1日-2018年6月30日的发病数,利用SPSS 22. 0建立多层感知器神经网络模型,变量选择"发病数",分类变量选择"发病年""发病月"指标,协变量选择"前1周同期发病数",比较实际发病与预测值的差异,用平均误差率(MPAE)对模型进行检验。结果构建的神经网络最优模型为24-9-7-1,即模型输入层含24个神经元,隐藏2层,分别含9个和7个神经元,输出层含1个神经元。手足口病预测发病数的自变量重要性依次是"前1周同期发病数""发病月""发病年",模型预测值与实际值相关系数r=0. 99,平均误差率(MPAE) 13. 37%。结论多层感知器神经网络模型预测手足口病发病情况有一定的可行性,能提供预测指标的重要性权重,"前1周同期发病数"是预测较好的指标,但预测指标合理选择和结果准确性方面有进一步提高的空间。

Objective To study the application of multi-layer perceptron neural network in the prediction of HFMD and provide the basis for the prediction of HFMD. Methods The date of HFMD cases from May 1,2008-June 30,2018 in Zhejiang was selected,the multi-layer perceptron neural network model was built based on SPSS 22. 0,the variables were selected as"incidence number",the categorical variables were selected as"year of onset"and"month of onset"indicators,the covariates were selected as"the number of cases in the last week",comparing the difference between the actual cases and forecast as well as finally testing the model with the mean percent absolute error( MPAE). Results The optimal model of the neural network was 24-9-7-1,which was that the model input layer contained 24 neurons and the hidden 2 layers including 9 neurons and 7 neurons and the output layer contained 1 neuron. The importance of independent variables for predicting the number of HFMD cases were as the following in turn: "the number of cases in the last week","month of onset"and"year of onset"model prediction and the actual correlation coefficient r = 0. 99. The mean percent absolute error( MPAE) was 13. 37%.Conclusion The multi-layer perceptron neural network model is feasible to predict the incidence of HFMD and it can provide the important weight of the predictors."The number of cases in the last week"is a better predictor but there is room for further improvement in the reasonable selection of prediction indicators and the accuracy of the results.