Research And Application Of K-means Clustering Algorithm

In recent decades,the data mining techniques develops quickly.It helps a lot in representing the past development of data,forecasting the future trends and supporting for decision-making in business and enterprise.Cluster analysis is one of the important data mining methods and have important applications in lots of fields.K-means algorithm is the most widely used clustering algorithm due by its efficient and rapid.This paper focuses on solving the disadvantage that K-means algorithm is susceptible to the initial point and easy to converge to the local extreme.That will help K-means algorithm get used more and more widely.This paper counts the number of samples that whose distance to others is less than the average distance of class center to others.WE judge the class center's degree of optimitation by this number and the number of the class.The new clustering index is all the class' s average degree of optimitation.This paper gives an improved K-means algorithm by setting up a thrishold to separate the class.This paper also merges several classes which are close to each other.By this way,another problem of K-means algorithm is solved.The simulation experiment in this paper estimated the special number of samples,and contrasted the different factors affecting the new indicators.We also get the value of the DB index among the classes close to each other in different dimensions and the thrishold in improved K-means algorithm within the new index.Experimental results show that,the improved K-means algorithm effectively solves the disadvantage that K-means algorithm is susceptible to the initial point and easy to converge to the local extreme.Finally,this paper analysised the different characteristics of users by improved K-means algorithm and provided alternative marketing programs for the makers.