Forecasting Methods Of Road Traffic Injuries

Background:Road traffic injury has become a social security and public health problem to be reckoned with in the world. Without effective intervention, it will be the fifth cause of death in the world by 2030. Most of road traffic injuries occur in developing countries, and only about 10% occur in developed countries. The WHO estimated that compared with 1990, road traffic injuries would rise 80% in developing countries and drop nearly 30% in developed countries in 2020. China is the biggest developing country, and is one of the countries with the largest number of road traffic injuries. With the rapid development of economic, number of motor vehicle is growing rapidly. But the improvement of the traffic management and the development of road construction are obviously lagging behind, which lead to an increasing trend in road traffic injuries and the number of casualties.60% of the deaths because of road traffic injuries occur in 16 to 45 years old young people in China, making impact on labor productivity population. So it is necessary to study on epidemic regularity of road traffic injuries and its development trend. This study via the regression analysis method, Autoregressive Integrated Moving Average Model (ARIMA) method, the grey model method and Back Propagation (BP) neural network method, using the data of road traffic injuries, establishes predictive models. The forecast of road traffic injuries, and comprehensive comparison of these methods, provide the basis for road traffic injuries prediction method choice, and provide a scientific basis for generating strategy of preventing road traffic injuries and reducing the accident severity.Objectives:1. The establishment and use of forecasting methods of road traffic injuries.2. To compare these prediction methods comprehensively, and provide the basis for the choice of road traffic injuries prediction methods.Methods:Using regression analysis method, the ARIMA model, grey model and BP neural network to establish forecasting models based on data of 1951-2012 road traffic injury in China.Results:1. The forecasting model of regression analysis based on data of 1951-2012 road traffic injury in China is Y=-0.000164X3+0.015669X2-0.267856X+1.275373. The model and its parameters are statistically significant. The predictive value (95%CI) of mortality rate of 100000 people of road traffic injury in 2012 in China is 5.74 (4.02-7.46). The forecasting model of regression analysis based on data of 2006-2012 road traffic injury in China is Y’=-0.439X’+7.079 The model and its parameters are statistically significant. The predictive value (95%CI) of mortality rate of 100000 people of road traffic injury in 2012 in China is 4.01 (3.22-4.79). The actual value is within the 95%CI of predictive value.2. The forecasting model of ARIMA is ARIMA(1,1,0):Yt=eYt-1+0.325▽Yt-1+et. Residual of the model is listed as white noise, and the P value of Ljung-Box test is bigger than 0.05. The predictive value (95%CI) of mortality rate of 100000 people of road traffic injury in 2012 in China is 4.67(3.05-6.87), and it consistent with the observed value. The actual value is within the 95%CI of predictive value. The forecasting model of ARIMA model is SARIMA(1,1,1)(0,1,1)12, and the P value of Ljung-Box test is bigger than 0.05. The predictive values of mortality rate of 100000 people of road traffic injury in 2012 in China consistent with the observed values, and the actual values are within the 95%CI of predictive value.3. The forecasting equation of grey model is x(k+2)= 6.0778e-0073k, and the predictive value of mortality rate of 100000 people of road traffic injury in 2012 in China is 4.22 with small error compared with the observed values.4. The BP neural network model based on data of 1951-2012 road traffic injury in China: data in the past four years for the network input, the data in the next year is expected output current cycle, the number of hidden layer neurons for 6. It was used to predict the mortality rate of 100000 people of road traffic injuries in 2012. The absolute percentage error is 0.67%.The BP neural network model data of 2000-2012 road traffic injury in China:data in the same period in the past five years for the network input, the data was expected output current cycle, the number of hidden layer neurons for 5. It was used to predict the mortality rate of 100000 people of road traffic injuries in 2012. The mean absolute percentage error is 3.8040%.Conclusions:1. Models established based on data of 1951-2012 road traffic injury in China, the BP neural network model predicts the best effect, followed by ARIMA model, and the regression model predicts the worst effect.2. Models established based on data of 2006-2012 road traffic injury in China, the prediction effect of the Grey model is better than the regression model.3. Models established based on data of 2000-2012 road traffic injury in China, the prediction effect of BP neural network is slightly better than the SARIMA model.4. When the sample data is not obvious seasonal, Choose the BP neural network model with enough sample data, and grey model with less sample data.5. When the sample data is obvious seasonal, the SARIMA model and BP neural network all can be used to build up the forecasting model. The prediction effect of BP neural network is slightly better than the SARIMA model.6. In the forecast of road traffic injuries, as far as possible, considering various prediction methods, finally make a comprehensive comparison, and choose the better prediction accuracy prediction method for modeling prediction.