Application Of Bayesian Methods In COVID-19 Event Warning Based On Gibbs Sampling

Since the raging of the COVID-19 pandemic that started in 2019,China has gradually liberalized its social control after a three-year dynamic clearance policy.The distribution of social materials,allocation of medical resources,response initiatives of various industries,medical safety of susceptible populations,and control of social opinion are all being tested.Meanwhile,conventional epidemic data will be distorted after the liberalization,and most of the existing epidemic prediction models are based on very high-quality data.Therefore,it is crucial to capture the risk of potential new coronavirus outbreaks in advance when official epidemic data do not represent the existing epidemic trends well,and provide informed early warning for control and resource deployment at the societal level in advance.With the aim mentioned above,this paper analyzes the official epidemic data published by the National Health Commission and public opinion data represented by the Baidu index,and first establishes an exponential growth model for estimating the potential outbreak possibility of the epidemic.Under this model,the values of the exponential term coefficients can reflect the potential upward or downward trend of the COVID-19 epidemic.Meanwhile,this paper introduces a Bayesian inference method based on Gibbs sampling to estimate the uncertainty of model output results.In the empirical analysis conducted in Beijing,Shanghai and other cities,this paper firstly demonstrates that the exponential growth model based on the Bayesian approach can accurately predict the outbreak of the epidemic.Second,the paper analyzes the detailed development of several epidemic events using the official data source of the Health and Welfare Commission and the Baidu index data,respectively,and demonstrates the excellent performance of the early warning model combining public opinion data and the exponential growth model in terms of accuracy and noise resistance.Among a total of 34 epidemic events,the recall rate of the model based on the data of the Health and Welfare Commission is 79.41%,while the recall rate of the model based on public opinion data reaches 91.17%,demonstrating the excellent early warning coverage ratio of the proposed model.In addition,this paper also proposes a machine learning integrator-based epidemic warning model that combines logistic regression,support vector machine,random forest and XGBoost as base models to predict the outbreak status.The model uses the prediction results of the epidemic warning model based on Bayesian methods as the training labels of the model.In the empirical analysis,the integrator of each machine learning model occupies the best value of 7 out of 12 indicators evaluated,which outperforms any single machine learning model.At the same time,the models achieve good precision and recall in Beijing and Urumqi data,which indicates that the machine learning models can provide more accurate warnings of the epidemic with high credibility and accurate features of the training data,and the epidemic warning models based on Bayesian methods have the ability to provide training labels for the supervised learning-based machine learning models.In summary,the model proposed in this paper can achieve a better outbreak prediction effect.In addition to the outbreak warning and epidemic stage analysis of Covid-19 pandemic,it can also be applied to the detection of other periodic social events,such as influenza,traffic rush and other universal social events,providing a strong reference basis and countermeasure guidance for various industries.