Analysis And Prediction Of COVID-19 Cases Based On Machine Learnin

At the end of 2019,the novel coronavirus outbreak broke out and spread globally at an extremely fast pace.Compared with SAS and MERS in the early years,the novel coronavirus has higher variability,and many countries and regions in the world are affected by the novel coronavirus epidemic.The novel coronavirus pandemic is a serious global pandemic,affecting people’s normal life,causing huge social and economic losses around the world and threatening people’s lives and health.It is of great significance for the prevention and control of the epidemic in the future to study the development trend of the epidemic,reflect the changing trend and future trend of the epidemic,and provide an objective basis for the government departments to prevent and control the epidemic.In recent years,machine learning methods have been widely applied in predictive tasks.This article focuses on the impact of factors such as vaccine administration and population size on COVID-19 case data.It utilizes LSTM models and their variant models to predict the cumulative confirmed cases of COVID-19.Furthermore,ensemble algorithms are employed to combine these models,aiming to construct a model with lower prediction error and higher accuracy.The main research work is as follows:Firstly,based on actual datasets,the Pearson correlation analysis method is used to select 62 features,retaining the feature subset that shows a high correlation with cumulative confirmed case data.This process improves the generalization ability of the model.Secondly,on datasets that include the relevant features from Australia,Israel,South America,and Chinese Taiwan,LSTM models,GRU models,BILSTM models,and CNN-BILSTM models are employed to predict the cumulative confirmed cases.The results are then compared with the predictions from models that do not include the relevant features.Finally,the two models with the highest prediction accuracy in each of the four regions are combined using Bagging ensemble technique.The prediction accuracy of the ensemble model is then compared with that of individual models.This article evaluated the established model using MAPE and R~2.The experimental results indicate that incorporating 10 feature indicators with high correlation to cumulative confirmed cases leads to significant improvements in the prediction performance of the four models compared to not including these relevant features.The MAPE values decreased by at least 0.0036%,and the overall R~2 exceeded 0.90.Furthermore,compared to a single prediction model,the fused model created through Bagging ensemble showed a minimum decrease of 0.0001%in MAPE and an increase of at least 0.0002 in R~2,resulting in more accurate predictions.