Stock Prediction Model Based On Gaussian Process Machine Learning Method Research

Recent a few years one of the most active direction in machine learning has been the development of practical Bayesian methods for challenging learning problems. Gaussian processes have been presented as one of the most important Bayesian machine learning approaches. This paper, mainly describe the Gaussian process, and discussed how Gaussian process connected with machine learning areas, and the key point is using the Gaussian process of machine learning methods in the prediction of securities.In this paper, through the progressive approach method, starting from the Bayesian linear regression model, layer depth, and ultimately reveals how the Gaussian process of machine learning methods used in the regression analysis. The introduction of the Gaussian process is mainly achieved through the projection of the eigenvectors in the feature space and function space. Next, we introduce several commonly used covariance function, and we can draw the conclusion that the covariance function is the key of Gaussian process model. So it is necessary to describe some simple and familiar covariance function, and how to produce the new and complex covariance function from the familiar ones, and make them have the property of the isotropic, non-isotropic, smooth and variable cycle. The property of the covariance function is reflected by the structure and the value of the hyper-parameters of the covariance function. By observing the structure of the covariance function, many types of covariance function we can easily understand the meaning of the hyper parameters. We studied the problem to choose an appropriate type of covariance function, which is followed by discussion of the choice of covariance function is the model selection problem. Finally, we introduce how to use the Gaussian process of machine learning methods to predict the securities. One of the methods is based on time series Gaussian process machine learning methods, mainly used for short-term forecast. And the method of enhanced Gaussian process variable machine learning method, the method used in long-term forecast, which is simple and easy to implement. The data which we used are the U.S. Dow Jones index which is applied to short-term forecasts. And the American Soybean futures which contain the data of daily transaction price, trading volume, open interest and time. These data are used to enhance the description of the data. The use of these two different Gaussian process machine learning methods for regression analysis of the data by MATLAB matrix computing power to achieve the purpose of prediction, the experimental results is satisfactory.