Research On Forecast Of Time Series Based On Svm

Time series forecasting is one of the main research topics in artificial intelligence and data mining. How to create a mathematical model by observing a finite number of historical samples is an important work of economic activity. We have lots of time series forecasting methods, such as the traditional time series analysis and neural network methods, which get advantages in dealing with stationary time series, to some extent, does not achieve the hope for results. Statistical learning theory (SLC) focuses on the machine learning theory of small samples. Its core is to control the generalization learning machine by controlling the complexity of models. Supporting vector machine (SVM) is a method of machine learning based on VC dimension and structural risk minimization principle of the statistical learning theory. SVM has advantages in solving small sample size problems in practical applications, such as small sample, nonlinear, over learning, no-linear, high dimensional and local minimum point. These problems exist in many learning methods.This paper mainly discusses either Support Vector Regression Algorithm or the classical time series analysis which is better at the prediction accuracy. The exordium, elaborating on the background of the problem, the purpose and significance of the research, the present research status at home and abroad and major contents of the research. In Chapter 2, the basic theory and method of traditional time series analysis is given. Such as AR model, MA model, ARMA model and SARIMA model. The third chapter describes the basic theory of support vector machines, and then introduces the Support Vector Regression Algorithm in detail. In the forth chapter, we take the example of passenger traffic to analysis the characteristics of seasonal time series. Next we use SARIMA model to predict the future trend of it. Chapter 5 is the core of the article. In this chapter, we use SVR and SARIMA model to predict the future trend of the monthly sales of a drinking listed Company. The results show that SVR method can predict non-stationary time series efficiently and has higher prediction accuracy than classical traditional time series analysis methods. In the final chapter, we propose the method integrated many kinds of advantages of forecasting methods and hybrid predictive models will have a vast potential for future development.