Prediction Model For Infectious Disease Incidence Based On Wavelet Neural Network Optimized By Genetic Algorithm

Some known infectious diseases may cause severe harm to humans or a pandemic, so many national governments borrow public authority to assist health care system to closely monitor occurrence and subsequent development of these diseases to avoid spread of epidemic. These infectious diseases are referred to as legal infectious disease. Infectious Disease Prevention Act lists 39 types of acute and chronic infectious disease with high incidence, large-scale spatial epidemic spreading and severe harm as legal infectious disease, and classifies legal infectious disease as A, B, C according to spread way, spread speed and harm degree.China practices localized management system for epidemic surveillance data collection of infectious disease, and adopts Direct Network Report System of Infectious Disease from January 1st,2004. Responsibility reporting units at all levels report online directly, implement by way of auditing at the county level playing a main role and sharing data according permission at all levels, achieves network real-time input straight, real-time query and analysis of infectious disease report card. At present, China’s major infectious diseases surveillance data is legal infectious disease reporting data. Time series prediction and analysis based on historical incidence data, to accurately predict infectious disease incidence is an important prerequisite for scientific planning and decision-making of health resources.The study used Wavelet Analysis and Genetic Algorithm to optimize Back Propagation Neural Network to establish prediction model for infectious disease incidence based on Wavelet Back Propagation Neural Network Optimized by Genetic Algorithm, and validated by example to compare prediction effect of different prediction model to explore feasibility of prediction model for infectious disease incidence prediction, so that infectious disease prevention work be with guidance of scientific evidence.The study collected epidemic situation reporting data for January 2005 to December 2014 of national legal infectious disease published on National Health and Family Planning Commission Disease Prevention and Control Bureau official website and population data from China Statistical Yearbook, calculated national monthly reported incidence of B reported cases top 3 diseases, Tuberculosis, Hepatitis B and Syphilis as example validation object of prediction model, used 4 types of prediction model, Back Propagation Neural Network (BPNN), Loose Wavelet Back Propagation Neural Network (WBPNN), Back Propagation Neural Network Optimized by Genetic Algorithm (GABPNN) and Wavelet Back Propagation Neural Network Optimized by Genetic Algorithm (GAWBPNN), to predict national monthly reported incidence of the 3 infectious diseases on MATLAB, and evaluated each prediction model by comparing the prediction result of different prediction models.For Tuberculosis, Hepatitis B and Syphilis, built model respectively by using national monthly reported incidence for January 2005 to December 2013 to predicted national monthly reported incidence for January 2014 to December 2014 respectively, combined actual national monthly reported incidence for January 2014 to December 2014 respectively as reference value to verify the validity of prediction model. Evaluation index was Mean Square Error (MSE), Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE), of which dimensionless index MAPE≤10% means prediction model had excellent capability of accuracy prediction. By using Rank-sum Ratio to comprehensively evaluate the average rank (R) of each index for the prediction models, compared prediction effect of the four models.For 2014 national monthly reported incidence of Tuberculosis, prediction error of BPNN prediction model was MSE=0.2225, MAE=0.4275, MAPE=5.95%; prediction error of WBPNN prediction model was MSE=0.1000, MAE=0.2562, MAPE=3.58%; prediction error of GABPNN prediction model was MSE=0.1527, MAE=0.2793, MAPE=3.95%; prediction error of GAWBPNN prediction model was MSE=0.0981, MAE=0.2420, MAPE=3.56%. The prediction result indicated that MAPE of each prediction model was less than 10%,4 types of prediction model had excellent capability of accuracy prediction. Rank-sum Ratio’s comprehensive evaluation result showed that, according to a=0.05, overlapping length of 95% CI of GAWBPNN and WBPNN was less than 1/2 length of 95% CI, GAWBPNN and GABPNN, BPNN without overlapping, the difference was statistically significant, GAWBPNN prediction effect for 2014 national monthly reported incidence of Tuberculosis was better than WBPNN, GABPNN and BPNN.For 2014 national monthly reported incidence of Hepatitis B, prediction error of BPNN prediction model was MSE=0.0516 MAE=0.1785, MAPE=2.64%; prediction error of WBPNN prediction model was MSE=0.0229, MAE=0.1241, MAPE=1.92%; prediction error of GABPNN prediction model was MSE=0.0435, MAE=0.1642, MAPE=2.41%; prediction error of GAWBPNN prediction model was MSE=0.0196, MAE=0.1179, MAPE=1.79%. The prediction result indicated that MAPE of each prediction model was less than 10%,4 types of prediction model had excellent capability of accuracy prediction. Rank-sum Ratio’s comprehensive evaluation result showed that, according to a=0.05, overlapping length of 95% CI of GAWBPNN and WBPNN was less than 1/2 length of 95% CI, GAWBPNN and GABPNN, BPNN without overlapping, the difference was statistically significant, GAWBPNN prediction effect for 2014 national monthly reported incidence of Hepatitis B was better than WBPNN, GABPNN and BPNN.For 2014 national monthly reported incidence of Syphilis, prediction error of BPNN prediction model was MSE=0.0242, MAE=0.1246, MAPE=4.88%; prediction error of WBPNN prediction model was MSE=0.0105, MAE=0.0823, MAPE=3.16%; prediction error of GABPNN prediction model was MSE=0.0093, MAE=0.0746, MAPE=2.70%; prediction error of GAWBPNN prediction model was MSE=0.0054, MAE=0.0594, MAPE=2.32%. The prediction result indicated that MAPE of each prediction model was less than 10%,4 types of prediction model had excellent capability of accuracy prediction. Rank-sum Ratio’s comprehensive evaluation result showed that, according to a=0.05, GAWBPNN and WBPNN, GABPNN, BPNN without overlapping, the difference was statistically significant, GAWBPNN prediction effect for 2014 national monthly reported incidence of Syphilis was better than WBPNN, GABPNN and BPNN.In this study, prediction and analysis of 2014 national monthly reported incidence of 3 legal infectious diseases, Tuberculosis, Hepatitis B and Syphilis, showed that BPNN and its optimized nonlinear prediction model all had excellent prediction effect, but compared with BPNN prediction model, optimized BPNN prediction model such as WBPNN, GABPNN and GAWBPNN had obvious advantage. Moreover, Wavelet Back Propagation Neural Network Optimized by Genetic Algorithm (GAWBPNN) had the most excellent capability of accuracy prediction. Therefore, predicting infectious disease incidence by using BPNN and its optimized nonlinear prediction model had the feasibility, and as the current example validation result showed, Wavelet Back Propagation Neural Network Optimized by Genetic Algorithm (GAWBPNN) prediction model was more suitable for infectious disease incidence prediction. Because of such a strong ability of nonlinear mapping, self-learning and global optimization, Back Propagation Neural Network Prediction Model and its optimization achieved good prediction effect for the three types of infectious disease incidence, with good application and dissemination value, suitable for building prediction model of universality for infectious disease incidence prediction, deserved further research.