Research On Adaptive Clustering Algorithm Based On DBSCAN Theory

With the explosive development of computer and communication information technology, massive amounts of data are accumulated in many field, which hiding a lot of valuable information. So much useful information is acquired for effectively promoting the development of social economy. Data Minning technology was born due to this demand, which is one of the most popular research directions of high value for application in computer and information technology field. Clustering technology is an integral part of the application of data mining. As a very popular research content, its main goal is to use and analyze data relevant characteristics to identify categorized between clustering objects.This paper mainly studied the adaptive of clustering parameters in DBSCAN algorithm, as well as the incremental clustering method of grid density division. The specific research content are as follows:1. Introduces the basic theory of kernel density estimation, analyses the DBSCAN clustering method on the advantages and defects, uses non-parametric kernel density estimation theory to analyze the distribution features of data samples, and researches and improves a method to automatically determine the Eps and minPts parameters, which can select appropriate Eps and minPts parameters. The clustering parameters are no longer needed to specify, and the whole process can be accomplished automatically. Experiment and simulation results show that this method not only improves the efficiency of clustering, but also adaptively determine the parameters for the two clustering.2. Introduces the basic ideas of incremental algorithm, skillfully uses a mesh division method, and with the aid of the parameter adaptive method in preceding chapter, studies a dynamic incremental clustering method based on DBSCAN. This method does not like the traditional incremental algorithm when the sample data set is updated, clustering is also started. However, it's just scanning the target sample, obtaining content and information of each grid cell division, and finally respectively dealing with a finite number of grid cells divided, and mapping the clustering information to the final results. Experiment and simulation show that the algorithm can effectively deal with the problem of data update, high efficiency and low complexity, simple and effective.