Nonlinearity Characteristic Analysis Of Financial Market Based On Support Vector Machine

The theory of support vector machine is a professional theory of solving nonlinear classification and regression problems which avoid the hard establishing of the net structure and the problem of over-learning, under-learning and local minimum. It is known as the best method of solving the classification and regression of using a small amount of sample. The support vector machine (SVM) provides a new thought of solving nonlinear problem.With the explosion of modern science and technology as well as the development of socialized production and the degree of internationalization of the substantial increase, integration of the world economy is one of the basic characteristics of social development. The exchange rate as an important link of international financial relations, and play an increasingly important role. Therefore, the correct analysis and forecast for the fluctuations in the exch rate is great significance for monetary policy and investment decision-makingSupport vector regression machine is one of the outstanding methods dealing with regression problem. It is widely used in multi-input and single-output problems recently. The multi-output support vector regression machine (MSVR) is defined for actual application, which can deal with multi-input and multi-output problems. Due to the high relevancy of multi output, the model structure of multi-output support vector regression machine is complex, so improving the forecast precision and the estimation of its error has become more difficult.This paper presents a new algorithm of data dependent kernel function for MSVR, which could effectively improve the forecast precision of MSVR, the proof of the data dependent kernel function which belongs to the kernel function of Mercer is also given. The result is used in the exchange rate prediction. The experimental results indicate that the method is correct and efficient for improving forecast precision and performance of MSVR.The correlativity is existed among these multiple-output. If the multi-output regression problems are estimated by some single-output SVR algorithms, it will have side effects, such as low accuracy, difficulty in parameter selection and high time-consuming computing. Considering the basic theory of MSVR, this paper presents a new algorithm of bound of Leave-one-out (LOO) error for MSVR, which could effectively estimate the forecast error of MSVR, the proof of the bound of LOO error is also given. The result is used in the microbial fermentation experiment. The experimental results indicate that the method is correct and efficient for estimating the forecast error and the performance of MSVR.