Prediction of Hepatitis C Incidence Using Adaptive Correlation Entropy Weight Method and Multivariate Time Series Model
10.11783/j.issn.1002-3674.2025.05.001
- VernacularTitle:联用自适应相关熵权法和多变量时序模型的丙肝发病率预测研究
- Author:
Tianhua YAO
1
;
Xicheng CHEN
1
;
Yazhou WU
1
Author Information
1. 陆军军医大学军事预防医学系军队卫生统计学教研室(400038)
- Publication Type:Journal Article
- Keywords:
Feature selection;
Information entropy;
Hepatitis C;
Time series;
Machine learning
- From:
Chinese Journal of Health Statistics
2025;42(5):642-648
- CountryChina
- Language:Chinese
-
Abstract:
Objective Hepatitis C is a kind of infectious disease with great harm and strong concealment.Accurate trend prediction is an important measure to ensure accurate intervention.This paper aims to confirm the effectiveness of multivariate time series prediction method and Internet data and provide a better data and method basis for hepatitis C prediction.Methods The data of the monthly incidence of hepatitis C in Chongqing from 2011 to 2018 were included,including infectious disease incidence data and Internet prediction data.To screen important features,this paper introduces the theoretical basis of feature entropy,and proposes an adaptive correlation entropy weight method(ACEW)for feature selection through the steps of collinearity removal,directional evaluation and information content evaluation.After that,this paper constructed a multivariate time series model(CNN-BILLSTM-Attention)and carried out the characteristic performance test(including prospective evaluation and posterior evaluation)and model efficiency exploration.Results Prospective evaluation revealed that the variables selected by ACEW had low consistency with each other,and the weight distribution calculated by each variable was relatively equal.The posterior evaluation revealed that the feature set screened by ACEW could obtain the best prediction information in each model.In the exploration of model effectiveness,the overall performance of multivariate time series prediction model is significantly better than that of univariate model.When ACEW and CNN-BILSTM-Attention are combined,the MSE,MAE,RMSE,MAPE and R2 on the test set are 0.0223,0.0649,0.0771,5.9285 and 0.9156,respectively.Conclusion In the study of predicting the incidence of hepatitis C,data fusion and method improvement are studied in this paper.The improved feature selection method(ACEW)can provide an opportunity for the regulation of hepatitis C,and the multivariate time series prediction model can improve the performance of hepatitis C trend prediction,to effectively control and prevent hepatitis C,which has better public health prevention and control significance.
