The Infectious Disease Early Warning Method Based On Statistical Learning Theory

Among all the emergent security incidents in public health,the most serious impact is the outbreak of the infectious diseases.It not only affects people's normal outgoing,but also causes the psychological panic of the public,as well as the unset situation in all aspects economically and socially.The emergency and development of the early warning technology,however,becomes especially important to prevent the outbreak of the infectious diseases and take effective corresponding measures.Among many statistic learning theories,there are many methods used for predicting and warning,and some of them are good.This paper focuses on three kinds of commonly used statistic model,selects the tuberculosis in the infectious diseases as the object,through the analysis of collected data about tuberculosis,using 10-fold-cross-test way compare,so that we could get a better model in the early warning.First of all,the paper introduces the situation of the emergent epidemic incidents in recent years and its impact upon people's life.At the same time,it introduces what kinds of measures have been taken facing unexpected events in many countries,through analyzing the advantages and disadvantages of the early warning models adopted by different countries,finally,the paper selects three kinds of statistic models,represented by the tuberculosis,compares and analyses the performance of these three kinds of models.Secondly,it introduces the basic methods of three models which are structural equation modeling,artificial neural networks,and random forest.Finally,it carries on the empirical analysis.Using R software programming language,processes and analyses the data of tuberculosis from many schools in Dalian City.The conclusions are as follows,structural equation models can express the unobservable hidden variables using multiple observable indexes,and may well express the relationship between the hidden variables.But the limits of it are that it's a verifiable model and therefor different models will have different consequences.The nonlinear processing capacity of artificial neural networks is great,but it's easy to have a fitting phenomenon.Random forest model has a fast speed to process data,and doesn't have a fitting problem.Using the 10-fold-cross-test way to test Artificial neural networks and Random forest,the results showed that Random forest has higher and more stable fitting degrees.