Sequence Singularity Based On Improved K-means Clustering Of Securities Time

The time series are arranged in chronological order, and changed over time and the sequence of interrelated data. The method of analysis of time series constitut an important area for the data analysis. Time series contain a lot of potential information, knowledge and rules. Many academics have been studying the time series with various ways, and trying to obtain the hidden value for further understanding the formation and development of things.The outlier in time series usually can effect the trend of the time series, at the same time, it often indicates social public information to the influence of the stock market. So how to detect the outlier in time series effectively become a popular research issue in recent years. This paper based on the clustering of the outlier detection by the idea, and combine with the K-means clustering algorithm, and put forward an improved K-means clustering algorithm. The main feature of the algorithm is to use the idea of the clustering of the outlier detection as a selection of basis of the initial cluster centers in K-means clustering algorithm, this algorithm uses the inherent disadvantage of the K-means algorithm which is easy to fall into the local extrema and led to the result that the K-means algorithm can not find all of the outlier when the cluster number is k. So we changes the number of clusters k and makes the number of clusters K=n·k. Then we let K decreases to k in a certain step size, In this process we have repeatedly cycle to identify the outlier, so that we can ensure this algorithm will not always fall into the local extrema.At the same time, the algorithm introduces the concept of information entropy,and make it as the condition for the termination of the algorithm. Then we use the algorithm to detect the outlier in closing price and stock trading volume. And we make research and analysis on relationship between the outlier and information factors. We conclude that stock trading volume can explain the changes in the stock’s closing price, and the information factors are the cause of the outlier. The conclutions have a very important reference value for that investors predict stock movements and stock investment.