Forcasting ShangHai Stock Index Using FTS Model Based On SVM-Modify

Stock market index forecasting is a difficult task. Traditional stock index research method stays in experience judge or simple of data analysis stage. Main method has fundamentals analysis method, and trading indicators analysis method. This class analysis method can not use information efficiently, or users must have rich experience. General of investors can hard directly use the class method, especially new entered stock of investors often will was around various rumors or pseudo expert misled, caused economic loss, Out of experience is an important stock market index forecasting of the agenda. Big data technology has developed rapidly, many scholars to move the focus to the areas of data mining, econometric analysis, called the new darling of the data mining algorithms. In the wide range of algorithms, fuzzy time series model is one of the most efficient algorithms used to predict stock market. Many scholars try to use fuzzy time series model to predict the stock market’s future performance, based on the direction of the evolution and improvement of the model documents are concluded, by taking a new model, as well as deficiencies in previous literature presents future research directions. Paper of innovation main has three points, first one is to propose new method of dividing data interval, which can use more effective history data that contains much information; Second is to propose new formula to calculate data membership degrees, which reflecting different of importance of past data when is used; Third is on model forecast results of amendment, through SVM classification algorithm amendment model forecast of were mixed case, last of data results displayed the amendment method effect significantly.The main contributions of this paper are:First, this paper proposed a new method of interval division. In the case of the introduction of information entropy makes the interval division has been theoretical basis, information loss arbitrary interval division brought down to a minimum. This is not simply to use of information entropy to divide intervals, but further deal data after the division of the stock on the basis of the characteristics of the historical data, including further dividing the range of the interval which is too large, merging intervals which contains only a single number and so on. These data preprocessing steps can make reasonable intervals contain sufficient information without having to divide too much range. The quantity and quality of the range has been some improvement; the second is to propose a new membership function. This is a main improvement field, but the researchers in this area are not many, and they mainly focus on the trends in the number of counts or weighted. The paper points out that distance of each data can be based on the different importance; third is the application of SVM classification model proposed to modify FTS models. Simple FTS model’s limitations are growing, leading to the improvement of individual measures to enhance the accuracy of the model can not meet the requirements, so the introduction of other auxiliary prediction model is important to improve prediction accuracy.