Application Of Generalized Regression Neural Network (GRNN) In Forecasting Incidence Of Hemorrhagic Fever With Renal Syndrome

IntroductionHemorrhagic fever with renal syndrome(HFRS)in Asia is also known as the epidemic hemorrhagic fever(EHF),it spread by Hantavirus in the caused by a different virus types.On the basis of reliable scientific HFRS the epidemic forecasting,and targeted anti-rodent and vaccination measures to the realization of the scientific HFRS prevention and control of major guiding significance.Over the years,many scholars have tried a wide range of traditional forecasting methods to forecast the epidemic of HFRS.Although the current forecasting methods have advantages,but mostly concentrated in its causal relationship between regression model and the analysis of time series model,the model can not fully reflect the essence and the forecast of dynamic data structure and the inherent characteristics of complex, which lost a certain amount of information.Artificial Neural Network is a unique parallel architecture,adaptive, self-organization,associative memory,and more robust fault tolerance and the unique characteristics of information processing methods,adapted to the incidence of hemorrhagic fever with renal syndrome the highly complex,nonlinear and uncertainties, and in the practical application have achieved remarkable success.So far,the forecast is the most used BP neural network model,but for such a network in the forecast,there slow convergence and vulnerable to the shortcomings of local minima in the sample solution less noise and more problems,the result is not ideal;sometimes predict the outcome of the subjective factors influenced.Generalized Regression Neural Network in the approximation capability classification ability and learning speed share a stronger advantage than BP neural network,the final convergence in the network gathered more samples of the optimization of return is better Extrapolation;and the lack of sample data,also forecast better results;In addition,it can be dealt with instability data.Therefore,this study the use of traditional BP neural networks and general regression neural network forecasting models were established and have fitting and predictable performance comparisons.Materials and MethodsThe morbidity data is from the Centers for Disease Control in Liaoning Province, select Shenyang City in Liaoning Province from 1985 to 2003,the incidence of HFRS, which collected a total of 5,304 cases of cases,access to accurate and reliable information on the case;rat situation Select from 1984 to 2002 the annual spring and autumn in monitoring the collection of monitoring data points,including density and the rate of mice infected.Meteorological data from the Shenyang Municipal Meteorological Bureau,select the region from 1984 to 2002,the average temperature (℃),wet gas(%),precipitation(mm)and sunshine(hr).The meteorological data(including average temperature,relative humidity, precipitation and sunshine)and animal disease information(including density and the rate of mice infected)six indicators in 1984 as the first group of input samples,the incidence of HFRS in a sample in 1985 as the first group of output.Until the end of the day in 2002 the meteorological data and animal disease information as a sample of six indicators of the final group input,the incidence of HFRS in a sample of a group in 2003 as the final output.The last three samples will be as testing samples,randomly selected one sample (to be assessed)from former sixteen samples to find the optimal smoothing factor,and the rest as a training sample.The input and output data were normalized before analysis.Use neural network programming toolkit of Matlab7.0 construct the incidence of HFRS general regression neural network model and BP neural network model. Absolute error,the mean error rate(MER)and the coefficient of determination R~2 are selected to test the fitting and prediction.ResultsThe optimize smooth factor of GRNN is 0.35;the hidden layers of BP neural network is 6.From the fitting effect,the MER of GRNN and BP neural network are 25.42%and 25.55%respectively;their R~2 are 0.9438 and 0.9729.On the whole,the fitting effect is satisfactory,and the difference of the two neural networks is not very significant.From the forecasting effect,the MER between the two neural networks are 4.90%and 15.16%respectively.The MER of GRNN is less than the MER of BP neural network;their R~2 are 0.9897 and 0.9516.ConclusionsGRNN has fast learning speed,better ability to address instability data,the small sample forecast accurately,the network structure less susceptible to subjective factors and forecasts Results stability characteristics than BP neural network model in forecasting the incidence of HFRS,so GRNN is of good practical value in solving the complex factors affecting the prevalence of the problem,such as HFRS.