The Application Of Time Series Data Mining On Security Analysis

Computer and information technology has brought to human society dramatic changes with the development of half a century. With the expansion of the scope of human activity, the fast pace of life as well as advances in technology, people can access and store data more easily, which makes the data to increase exponentionally. Therefore, people fall into an embarrassing position, that is, the rich data and the poor knowledge. Data mining, as a new method of data analysis, is used in more and more enterprises. It is widely used in database marketing, customer relationship management, customer behavior prediction, market trends forecast and so on. It plays a important role in processing mass data and solutes the problem of"data rich"and"knowledge poor"effectively.With the rapid development of the securities industry, the securities industry is facing competitions. The requirements of accuracy and quantitative of data are very high when the dealer in securities make decisions. It is possible that data mining which is a tool for data analysis apply to securities industry. There are a large amount of data in stock market everyday. These data will not only affect the stock's trading activities, but also be the important basis of the investors who want to buy stock. There are some uncertainties as well as risk in stock market because of economic, political, interest rates, supply and demand. Some people who are lack of knowledge of stock are subject to the restrictions of time and space, so it is hard for them to long-term concern about the stock market, which makes the forecasts on stock market become a necessity. The current forecast on the stock market mainly in basic prediction and technology prediction method. In this paper, we will introduce a new method of stock prediction which is based on time series data mining.We analyze the characteristic of the securities market, find that the securities market has unitary data, multiplicity, the characteristics of data mining is in line with the requirements of a large number of sufficient data and knowledge of decision, therefore point out that data mining is feasibility applied to security analysis of the data.Time series data mining is an important part of data mining, which is a relatively new area in data mining area relative to some area. Time series prediction include Long term trend, Cyclical component, Seasonal component, Irregular component. We can finish forecasting, modeling, characterization through the analysis of the time Series.Time series data are of high dimension, noise and various distortions such as amplitude scaling, stretching or compressing in the time-axis. and the stock data is a continuous quantity. It is difficulties to deal with such huge data, so we try to process the time series of stock data. The method distills the key point as the boundaries of sub-time series from the time series which is processed. We can choose the threshold value according to the conditions, different threshold determine different accuracy. When the threshold value is bigger, the segments you get are fewer, the compression ratio is higher and the details of the time series is less. When the threshold value is smaller, the segments you get are greater, the compression ratio is lower and the details of the time series is more. Then we can fit each subsection liner fitting function by using maximum likelihood function. The method not only retains the original time series characteristics and the position of each sub series but also compress the length of time series and the noise.Time series similarity search is an important aspect of time series study. There is a description of his problem, that is to identify with the most similar sequences to the given time series from a large time series database. The two time series are compared to find the offset and then we can determine whether the two time series is similar or the degree of similarity. At present, time series similarity search include Euclidean distance, dynamic time warping and so on. In this paper, we draw fitted characteristic vector of the segments, that is, the slope of the straight line tangent value. we use time series similarity search to find the trend which is similar to historical trend to carry stock prediction point.In this paper, the stock prediction system is completed with java and the development platform is eclipse. At the same time, we use jexcelapi to read and write stock data which is in excel format and use JFreechart to chart so that the results can be displayed intuitive. The historical data of stock are get by stock software. We use this program to analyze and predict the historical data in order to find the trend which is more similar to the historical trend. We expect to analyze current trend by historical trend.In this paper, we analyze current trend by historical trend and use time series data mining in stock market prediction. This method is of a certain reference value for mass of shareholders.