Research And Implement Of The Tuberculosis Infectious Disease Prediction Model Based On Machine Learning

Tuberculosis caused by mycobacterium tuberculosis can spread from person to person.Among all kinds of infectious diseases in China,the incidence of tuberculosis occupies the forefront.Although there is a model that can predict its future development trend,it takes into account very few factors and the prediction accuracy of this model is relatively poor.With the continuous development of deep learning in the field of machine learning,its application has been very extensive.Therefore,using the deep learning algorithm to predict the tuberculosis infection can be a direction of the research and application.The intensity and development trend of pulmonary tuberculosis in the regions of sichuan province were studied by deep learning technology.At the same time,combined with relevant early-warning technology,this thesis studied its possible epidemic scale in the future.And early warning and notification of abnormal conditions will be given.Firstly,this thesis studied the factors affecting the tuberculosis epidemic,then obtained relevant data through existing data collection methods,and standardized data from different data sources through data preprocessing technology.Meanwhile,it also studied the relationship between influencing factors and pulmonary tuberculosis infection from an epidemiological perspective.Because pulmonary tuberculosis had the temporal and spatial characteristics and ARIMA model commonly used for the prediction of infectious diseases had some problems,a gcTrans model based on the characteristics of GCN and Transformer in deep learning was proposed to slove the problems of ARIMA model.At the same time,in order to illustrate the effectiveness of the model,Transformer's encoder was used as another comparative model in this study.Then RMSE,MAE and R2 were used to evaluate the effect of these models.Finally,the prediction effects of these three models at multiple time points in the future were compared.The proposed model not only solved the problems of complicated steps of ARIMA model construction,a factor considered and inability to extract spatial features,but also solved the defect of too smooth features on each node in the graph after being processed by GCN.Through the above experimental comparative analysis,it is found that gcTrans model is better than other models in prediction accuracy.And the proposed model has long-term prediction ability.Finally,a prediction and warning system based on above research was designed and implemented.The system makes it easy for public health staff to observe its future trends and detect anomalies earlier,so that measures can be taken to control development of tuberculosis and minimize the loss brought by tuberculosis to society.