Research On Prediction Of Cardiovascular Death Based On Neural Network And Grey Model

Cardiovascular disease is the most common disease that seriously harms human health,and the related deaths are increasing year by year.The meteorological factors and air pollutions importantly affect the occurrence and development of cardiovascular disease.Further understanding the risks of cardiovascular disease and actively exploring the onset factors would contribute to reducing the deaths caused by the disease.The predication model can predict the cardiovascular deaths by considering the overall effect of meteorological factors and air pollutant on the cardiovascular disease.However,the traditional prediction model has the limitations of low accuracy and ineffectiveness,mainly due to the multiple affecting factors for the cardiovascular deaths and the complex nonlinear couplings among them.This thesis collected the datasets of meteorological factors,air pollutant and cardiovascular deaths from January 1 to December 31,2013 in Shenzhen City.Based on the BP neural network and GM?1,1?model,the prediction models were constructed to predict the cardiovascular deaths,and the prediction accuracy was further improved by optimizing the models.The main research content is as follows:?1?Choosing the suitable indexes.We chosen the four meteorological factors of average temperature,average humidity,atmospheric pressure and wind speed,and the six air pollutants of SO₂,NO₂,PM₁₀,PM_2.5,CO,and O₃,as the indexes which were used to be the input variables in the predication models.?2?Data preprocessing.We first filtered the raw datasets and obtained the 365 valid data.Considering the nonlinear correlations among datasets and the magnitude differences among the affecting factors,we performed the correlation analysis and normalization on the raw data to eliminate the mapping distortion.?3?Constructing the cardiovascular deaths predication model based on the BP neural network.The reasonable neural network structure was determined according to the input and output variables,in which the model codes were constructed with MATLAB software to predict the cardiovascular deaths.The mean absolute percentage error,acceptability,and predication accuracy were proposed to evaluate the predication model.?4?Constructing the cardiovascular deaths predication model based on the GM?1,1?model.Considering the temporal dynamic process of cardiovascular deaths varying with the time,we choose the history data of cardiovascular deaths as the research target and used the sliding-time window method to construct the GM?1,1?predication model.?5?Constructing the optimal predication models.First,the dimension of input variables was reduced by the PCA to decrease the complexity during training process,such that we constructed the optimized PCA-BP neural network model.Second,we combined the GM?1,1?model and BP neural network to construct the composite grey neural network model,which effectively overcomes the limitions of the two separated models and significantly improves the predication accuracy.Finally,we compared the predication results among PCA-BP neural network model,series grey neural network,parallel grey neural network model,BP neural network model and GM?1,1?model.Our results suggest that the BP neural network predicating the cardiovascular deaths has certain practical significance.While considering the historical data of cardiovascular deaths,the mean absolute percentage error of the GM?1,1?model is 20.70%.After decreasing the dimension of the input variables by PCA,the predication accuracy in PCA-BP model increases by 7.57%comparing to the BP neural network.In the series and parallel grey neural network models,the mean absolute percentage error decrease to 18.41%and 14.98%,and the predication accuracy are 82.30%and 87.59%,respectively,which significantly are larger than those in the separated GM?1,1?and BP neural network model.Importantly,the parallel grey neural network model has the highest predication accuracy.Our results not only would help the patient to avoid the danger environmental factors,but also could provide technical guidance for the further predication of other diseases.