Operating System In Time Series Analysis

Data Mining attracts more attention in both academy and industry in recently year. The data size of many fields are increasing tremendously whereas useful information extracted from this data is necessary. Time Series analysis is one important part of data mining, which has it wide application in realities. For instance, the analysis of stock, the analysis of cardiogram, sales of some product are examples of time series analysis.In this thesis, some effective methods for time series analysis are discussed. The properties and limitations are introduced, especially the theory and implementation of ARIMA model.Only the analysis result of time series could not meet the requirement of some practical systems. There are usually hundreds of or thousands of data item in application systems. It is impossible for managers to analyze all the data, Therefore a clustering algorithm for the data items is desirable. Features of time series are first extracted and then clustering algorithms are performed according to the features. In this paper, we proposed the method for feature selection according to specific predicting models. Time series that contains one or more observation points could be represented using features of low dimension while keep most of the information at the same time.We applied our algorithm in a practical system: there are lots of historical data that are necessary to be processed to provide decision support. However, useful information is usually hidden in complicated data. According to the method we proposed, we built up a decision support system. In this system, analysis and clustering of each time series data are provided with some decision recommendation.