A Study On Cluster Analysis Of Comprehensive Stock Index Data Based On Kmeans

Since the stock market appeared,people have been studying and exploring its inner laws.With the development of data mining technology,data mining has been introduced into the research field of stock analysis.Clustering is a common method used in stock analysis.In these years,researchers use cluster analysis to get the useful infromation that cannot be obtained by traditional analytical methods.Among all clustering algorithms,Kmeans clustering algorithm is widely adapted to various types of data.Therefore,we will use Kmeans as the basis algorithm of clustering analysis.However,when using Kmeans to analysis stock data,we found that Kmeans does not always fit the stock data.It means the different clusters in the clustering results cannot be well separated from each other.After analyzing,we believe: the reason why Kmeans can not get effective clustering results is that it will lost important information when using Kmeans to analysis stock data,which other researchers seldom consider.It is well known that when investors study stock data,they will pay attention to the relationship between the different curves formed by the same technical index in different parameters.However,if we use Kmeans to analysis the stock data,the relationship between these curves will be lost.Therefore,we focus on how to retrieve the lost information in this paper.In order to solving the problem,we propose two new indicators to retain useful information.And then we verify the feasibility of our proposal algotihm by experiments.Since the two new indexes are computed in the data preprocessing stage,we can regard the proposal method as a special data preprocessing method.In order to verify the effectiveness of the proposed data preprocessing method,we set multiple contrast experiments.The experimental results show that the proposal method significantly improves the efficiency of the Kmeans clustering algorithm,where Kmeans can be more effective to find the most suitable parameter K which means the number of clusters and clusters in clustering results have higher density and lower dispersion.Overall,the proposal data preprocessing method makes the Kmeans algorithm more effective,in dealing with the stock index data.Meanwhile,the experimental results of multiple contrast experiments show that the proposal method is widely applicable to the stock data.