The Application Of The Neural Network Model In The Prediction Of Acute Myocardial Infarction (mi) And A Comparative Study Of Model Prediction Ability

ObjectiveCardiovascular disease harms to human health seriously worldwide. In recent years, many studies have shown that incidence and mortality has been increasing in developing countries. Several studies have explored risk factors for myocardial infarction and predict the incidence rate. Statistical models need to be built for disease incidence rate prediction, and conventional statistical model predictive ability is limited. We hope to find the mathematical model which is able to analyze more complicated non-linear relationship between variables, and provide a reference for the diagnosis and prevention of acute myocardial infarction in China population. The neural network model is one computing system developed on the basis of the simulation of the human brain tissue, and it is network system constituted by a large number of processing unit through an extensive interconnection. The neural network model has the basic characteristics of the biological nervous system, and it has nonlinear mapping ability, learning ability, adaptive capacity, fault tolerance, and associated storage function, so it is an important model in data mining methods.The purpose of this study was to build a Logistic regression model, BP neural network model and Elman neural network model, and conventional statistical methods combined with the method of the neural network model applied to the prediction of acute myocardial infarction. It was expected to be able to improve the capability of the disease prediction.Methods All the variables consisted of conventional variables and the SNP locus variable from acute myocardial infarction epidemiological survey data in China population. The conventional variables were divided into qualitative variables and quantitative variables, and description of the variables and univariate analysis are applied. About gene SNP loci variable, we calculated gene and genotype frequencies, and finished Hardy-Weinberg equilibrium validation, trend test and SNP loci haplotype area construction.We constructed three kinds of statistical prediction model, conventional logistic regression model, BP neural network model, and Elman neural network model, calculated the area under ROC curve back inputting data to model, and compared the primary prediction accuracy of three models. Then we divided all the data into training and validation sets by random sampling method, and rebuild three models to evaluate the generalization ability. Repeated sampling method was used to compare three models prediction accuracy. Finally, we randomly simulated data, and taking into account the difference between the continuous variables and discrete variables in the model, so the random simulation was divided into two parts:the first, continuous variables were statistically significant; the second, discrete variables were statistically significant. We established models to evaluate adaptability and stability to different variables.ResultsAfter random sampling, the data was divided into the prediction data set and validation data set. We fitted3models, and compared the predictive ability between different models. The results showed that in4different proportions of validation data set,10%-40%, the area under ROC curve of BP neural network model were-increased by4.5%,3.1%,3.3%and2.9%, compared to Logistic regression model, respectively. The area under ROC curve of Elman neural network model were increased by4.2%,2.1%,2.9%and1.4%compared to Logistic regression model. The result fitted by20%and40%population as a validation dataset was not statistically significant. Differences of area under ROC curve between BP model and Elman model in four different proportions of validation data set were0.2%,0.9%,0.4%and1.6%. and there was not statistically significant. BP neural network can significantly improve the generalization ability of the model model compared to conventional logistic regression model.Random simulation data results showed that in the first part, simulation of continuous variables were statistically significant, and the predictive capacity of the three models was high. In second part, discrete variables were statistically significant, and in4different proportions of validation set,10%-40%, the area under ROC curve of BP and Elman neural network model increased by3.2%,2.9%,3.2%and3.1%, compared to Logistic regression model, respectively. Two kinds of neural network model predictive capacity were significantly higher than the Logistic regression model. Elman model and BP model were no significant difference.ConclusionThe results of this research showed that the application of BP neural network and Elman neural network model had high predictive capacity, faster computing speed, well stability and strong capacity to resolve complex non-linear relationship. The prediction performance of the neural network model was higher than Logistic regression model, especially in the little sample size situation, more discrete variables, and complex nonlinear relationship in the research data. It clearly showed that advantage and rationality of neural network. The application of these two kinds of neural network in prediction and evaluation in the field of heart disease epidemiology will have a better practical value.