Singular Point Of Securities Based On The D-nearest Neighbor Clustering Time Series Studies

Time series is a kind of expression form of complex data which can effectively describe society, economy, science and engineering fields. Through analyzing and studying time series, we can find and reveal the internal rules of objective data describing with change and development, which has a very important significance for the development of social economy and technology.Outlier detection is that finding interesting small pattern in the large data set, its study object is a tiny part of data from most of the objects. In recent years, the outlier detection research along with large scale data mining in depth has become a focus of research. There are wavelet analysis-based, density-based, distance-based and clustering-based methods, and so on. With the reason of the turning point of short time trend reflecting the movement trend of time series, if analysis the special points,we can get lots of important potential information. For the research on outlier of stock time series, in this paper, the thoughts of short-term trend clustering are introduced for researching the outliers in time series with cluster analysis. A method detecting the outlier of time series of stock prices based on d-nearest neighbor clustering is proposed. Study on the outlier of time series in stock market is dealt with in three stages. In the first stage, the breaking points where the short-term trend changes are used to be observing points. And then, clustering the short-term trend. In the second stage, the d-nearest neighbor clustering method is introduced to detect the outliers. In the third stage, the short-term trend of outliers is used for linear trend modeling with the least square method forecasting the future trend of stock time series, so as to guide decision.