Research On Multi-valued Function Recognition Algorithm Based On Clustering Method

The 21 st century is a century of information explosion,with the development of Internet technology world communication is becoming more and more convenient,in scientific research,bioinformatics,the application of Internet and e-commerce,and many other applications,data volume is growing at breakneck speed,in order to analysis and use of these huge data resources,must rely on effective data analysis technology.Big data application is a process of mining effective information from big data,providing decision support for users and realizing the value of big data by using the method of data analysis.In the process of big data processing and analysis,we will often find that the data from the same source will show multiple different function images in the same coordinate system,which will make the same variable correspond to multiple values in the data prediction and analysis,and we call it a multi-valued function here.How to identify the functional correlation in the big data set,how to extract the information of these multi-valued functions from the big data set,and how to perform better function fitting of the extracted multi-valued functions are all problems to be solved at present.This paper mainly studies the recognition and extraction of multivalued functions based on clustering method,and proposes a new solution for these three points.The main research results of this paper are as follows:1.The concept of multi-valued functions in large data sets is proposed.The mathematical concept of a multivalued function is a binary relationship in which each input corresponds to at least one output,and some to more than one output.In the process of big data processing and analysis,will often find the same source of data in the same coordinate system will show several different function image,or before a certain node is a function with a appeared after the node branch function,it will exist in the analysis of data to predict the same variable will correspond to the phenomenon of multiple values,we call it here in large data set of many-valued function.2.Two kinds of MIC calculation methods based on random Windows on data were proposed to preliminarily judge whether there is correlation.The first algorithm,W_MIC algorithm,uses the result of the maximum mic value of local data to determine whether the data has the value of functional relationship mining.If there is strong correlation in the local part of the data,we think the data is value mining.The second algorithm,A?MIC algorithm,uses multiple random windows to get the mean value to estimate the MIC of the global data.The result is similar to that of the global data.The above two algorithms are used for different purposes,which will greatly save the cost of computing time.3.An MKSR algorithm for multivalued function recognition based on spectral clustering is proposed.Use of the advantages of spectral clustering on the graph cut,combined with K-means method and gauss-Newton iteration algorithm proposed MKSR algorithm,first using the MIC value calculation of the data,then using the method of K-means clustering for large data set to initial data center,the classes in the data center of spectral clustering divides tag,finally respectively to different markup with Gauss-Newton iterative algorithm to find the corresponding functional relation,so as to realize the recognition of many-valued function and output.