The Key Technology And Its Application Of Time Series Data Mining

Data mining is the observed for analysis in order to find some new rules and new knowledge that is useful and easy to understand in the dataset. As the important class of the observation datasets, the time series is a collection of data in chronological order such as the stock data, ECG data, seismic data,and widespread in financial, medical and architectural applications. Time-series data have the characteristic of noise interference, high-dimensional features and updated quickly, and the approach for time series analysis is often used on statistical knowledge in previous studies. Due to the complex features of time series, the time series analysis for a large number of scholars has hindered. At the present stage, with the exponential increased of data, it is very necessary to analysis time series. Data mining has also attracted a large number of research scholars to study and try. In the dissertation, we carried out the time series of re-description method, time series similarity measure and time series data mining tasks including clustering, classification and outlier detection analysis. The main ideas are listed as follows:(1) A new classification method based on area extreme point of time series is proposed. By analyzing the characteristic of extreme points on the time series, an extreme point extraction strategy of time series is proposed. Based on this strategy, by using dynamic time warping distance measure to measure the similarity between the extreme point series, then AEPSM will be applied to the time series classification algorithm. The method can be well fitted the original time series and also can be to achieve efficient data compression and to obtain good results in terms of classification accuracy.(2) A new clustering method based on area extreme point and symbolic representation of time series is proposed. By analyzing the characteristic of extreme points and symbolic representation of time series, we propose a method that uses symbol method to describe the extreme points of time series, getting a series of symbolic sequences. Then, by using dynamic time warping distance for symbolic sequences to get the similarity measure, and finally AEPSSM will be applied to the time series clustering algorithm. This method can describe and measure the time series effectively, and can also obtain good results in terms of clustering accuracy.(3) Two outlier detection methods of time series are proposed. By analyzing the time series outlier detection method and applying LEP and LEPSAX to outlier detection algorithm, OD_LEP and OD_LEPSAX is proposed. Results which get from experiment by using synthetic data and real data show that the two detective methods have high accuracy.