A Study On The Evolutionary Clustering In Data Mining

Data mining technology is an effective means of obtaining useful informationand knowledge from massive data, and it is a multi-discipline integration ofcrystallization that has important practical value and broad application prospects. Datamining technology is flourishing while also facing challenges that traditional datamining techniques can only handle static data in the database, but in the actualapplication data is often dynamic, evolutionary data is such a data which datadistribution changes as time changes, the traditional data mining techniques can nothandle this problem. Thus, the study of specialized algorithms to deal withevolutionary data is very necessary.Evolutionary clustering is the key and difficult point in learning on evolutionarydata, In this article two exponential decay evolutionary clustering frameworks wasproposed. According to the selected prototype algorithm and smooth regularizationterm meaning we got four algorithms: KM-ED-PCQ algorithm, NC-ED-PCQalgorithm, KM-ED-PCM algorithm and NC-ED-PCM algorithm. This paper studiesthe following three aspects:First, the article briefly describes the data mining and traditional cluster analysistechniques, and then study evolutionary clustering, describes the features of theevolutionary data itself, the evolutionary clustering research status, commonevolutionary clustering method, etc. and carries out a compared analysis betweenexplicit modeling and smoothing the regular clustering methods.Secondly, In order to improve the smoothness of the evolutionary clustering, thepaper increases the time regularization, considering the time regularization impact atdifferent time on current time clustering, the exponential decay is applied. Dependingon the difference of time regularization meaning get two evolutionary clusteringframeworks ED-PCQ and ED-PCM and then apply K-means and spectral clusteringrole to the above two frameworks respectively getting four practical evolutionaryclustering algorithms.Finally, by carrying out experiments on Gaussian data sets and KDD-CUP99datasets verify the proposed algorithms is feasible and effective, and get the relationship between the time regularization changes in the number and degree ofclustering items, which brings the actual convenience for calculation.In summary, the two exponential decay evolutionary clustering frameworkproposed in this paper can handle of the clustering on evolutionary data effectivelywith a strong theoretical and practical significance.