Integrated Price Prediction Research Based On Support Vector Regression

In the process of economic development, the price reflects the change of the relationship between the supply and demand of commodity. The price and market demand is closely related. The price is one important means to achieve national macro-control. So price prediction is very important. Price prediction is a premise of sales prediction, profit prediction and financial prediction. The main idea of price prediction is analyzing the price-related information and data by using scientific methods to judge the future price changes. Price prediction can judge the trend of the future price changes of a commodity. It also can make the number judgment of a commodity by using a prediction model.Now, price prediction model mainly includes the grey system prediction model, the neural network prediction model and the time series prediction model. The main idea of grey system prediction model is founding the development pattern of the system by using existing data to establish the grey model, and then predicting the future price scientifically. The common grey system prediction model is GM(1,1) model; The main idea of neural network prediction model is selecting the appropriate neural network, and training the network by using existing data, and then getting the prediction results by giving input to the trained network. The common neural network is BP neural network. The main idea of time series prediction model is analyzing the time series which is formed by existing data, and establishing the appropriate time series model, and the predicting the future value of time series, that is to predict future price. The common time series prediction model is ARMA model.Support vector regression is an important application of support vector machine on regression problem. Support vector machine is a classification approach based on the statistical learning theory. It can achieve the classification of linear data and nonlinear data by searching the best separating hyper plane. For the nonlinear data, we must map the nonlinear data to a higher dimensional space by using kernel function, and then search the best separating hyper plane in the new space. An important concept of support vector regression is the loss function which allows the error within a certain range between the true value and the prediction value. Because of the presence of loss, we add the loss into the objective function by using a penalty factor C to balance the loss. An important step of establishing the support regression model is the selection of kernel function and the determination of the parameter of the loss function and the penalty factor C.Most of the existing price prediction methods are using a single method. But the problem of price prediction is very complicated. And there are shortcomings in the single prediction model. The major problem in the price prediction is how to integrate the prediction models effectively and get more accurate prediction result. To solve this problem, this paper advances a prediction model based on support vector regression integrating grey system, neural network and time series. The establishment of an integrated model consists of two steps:the first is establishing the grey system GM(1,1) model, BP neural network model and univariate time series model by using existing data, then predict future price. We can get three prediction results; the second is predicting the future price by establishing the support vector regression model. Specifically, we should establish the support vector regression model by selecting the kernel function and determine the parameters of loss function and the value of penalty factor C. Then get the prediction results.The actual data used in this experiment is the national average corn price in January 1996 to June 2009 which is provided by the official website of the Chinese livestock industry. Establish predict model by using the data in January 1996 to December 2008 which is training data. And Test the effect of predict model by using the data in January 2009 to June 2009 which is test data. Then compare the predict models. The experiment shows that the 6-month average relative error and root mean square error of integrated price prediction method based on the support vector regression which this paper advances are 1.99% and 0.0348(grey system GM(1,1) model:4.85% and 0.0742, BP neural network model:5.26% and 0.0902, univariate time series model:4.4% and 0.0637). Therefore, not only from the point of view of image, but also from the point of view of mathematical error analysis, we can draw the following conclusions:for the data prediction, the integrated model based on support vector regression is superior to the other three basic models.The integrated prediction method advanced in this paper has been applied in the issue named the research and application of the grid of digital agricultural knowledge. The issue is the National High Technology Research and Development Program, the 863 research topics. Our research group has developed a system named Agricultural Knowledge Grid. And the integrated prediction method researched in this paper is used in the price prediction module of the system.