Time Series Singularity Trend Direction Research

A time series is a huge complicated ordered data set which contains a lot of useful potential information and laws of social development. How to use the previous time series effectively to reveal the laws behind these data have been attracted much attention by scholars from home and abroad in recent years. Time Series Data Mining can quickly and effectively abstract the useful information from the many history data. It is often applied to mine the frequent patterns in order to find some laws. Abnormal data will be abandoned as noise because unnecessary. Abnormal data does not occur very often, but it can provide more valuable information.When modeling via stock time series, Abrupt Change Points or smallest proportion of isolated singularity is called as anomalous diffraction spots of stock time series. Because information is a decisive factor in stock market for price change, anomalous diffraction spots of stock time series marked certain information have affection for stock time series. Some data mining algorithms try to mitigate the effect of anomalous diffraction spots, but the tread for anomalous diffraction spots has an effect on time series show some characteristics and have certain information for investment. Overall, mastery information affection and find the relations between the anomalous diffraction spots of stock time series and information is very important for stock forecasting and investment.Abnormal data detection as the branch of data mining has gained more and more attentions in recent years. These algorithms can be roughly classified into three kinds: method based on distribution; method based on depth; method based on cluster; method based on distance and density. Because the anomalous diffraction spots of stock time series caused by information, it is local for stock time series. The mining algorithms based on density can detect the anomalous diffraction spots, but it can't be used directly for order time series. This paper accordance with the characteristics of anomalous diffraction spots, using slip window segmentation of time-series data detects local isolated singularity. Efficient Market Hypothesis shows market price can reflect given information clearly. Information of stock market has classified into good news and bad news. Different stock may show different reaction against the same information. So the judgment of the direction of isolated singularity can show the influence of information for stock time series. Based on local density detection, we designed a method of detect the direction of isolated singularity. Results of the analysis results of real data verify the efficiency of the method. At last, we categorize the direction of isolated singularity as forward and reverse direction, analysis the tread of different isolated singularity. And studies show that direction of isolated singularity affect the stock time series.