Financial Time Series Clustering Based On Dynamic Time Warping

With the development of social economy and information technology,the amount of data in time series grows more and more.The potentially valuable knowledge in the time series database discovered by data mining techniques has reached more and more attention,and research result have been successfully aplied in various fields.Time series similarity measure is a measure of similarity between time series,its measurment results can be used for classification,clustering,sumilarity search and orther time series data mining tasks.Time series clustering is one of the important mining tasks in the field of time series data mining.Different time series clustering methods can excavate different implicit information.Based on the time series as the research object,to explore the time series similarity measurement and clustering methods,the methods can be more effectively used in time series data mining,and then obtain the valuable information and knowledge.The main contents of this paper are as follows:(1)From the point of view of the numerical distribution and the characteristics of morphological fluctuation,a similarity measure based on numerical symbols and morphological features is proposed.The new method can fully reflect the time series numerical distribution and morphological features,and effectively improve the time series similarity of the measurement effect.(2)In order to solve the problem that the traditional clustering methods usually need to determine the number of clusters,and can not fully reflect the overall spatial structure of the time series and the relationship between each other,a new Clustering method for time series of label propagation based on centrality is proposed.The method can automatically cluster without specifying the number of specific clusters,and construct different network spatial structures according to different parameters.The number of clusters can be adjusted accordingly to improve its performance in time series clustering.(3)Application of dynamic time bending and time series clustering in financial field.On the one hand,based on the dynamic time warping and the classical timeseries clustering method,this paper makes a further exploration in the financial field.Through the study of stock linkage,we can draw the implicit information of stock time and provide some reference for investors to choose the portfolio.On the other hand,using new similarith measurment and clustering method to cluster Shanghai and Shenzhen 300 index and select trailing constituent stocks,and establish the minimum error tracking model to determine the optimal weight of the constituent stocks in the portfolio.Under the different investment ratio constraints,the new method determines the constituent stocks to more accurately simulate the underlying index,and can meet the different investment preferences of the investor investment requirements.In this paper,the validity of the new method is tested by numerical experiment.The performance of the new method is compared by the correlation method in the field of time series data mining.The new method further improves the time series similarity measure and clustering research,expanding the application of time series data mining in the financial field.