Weight Vector Based Multi-scale Clustering Algorithm

Data Mining is the process extracting knowledge which is potential, unknown, anduseful from extreme large database containing relatively complex structure. It is also knownas Knowledge Discovery in Database. Clustering, as an important field of Data Mining hasbeen considerably developed, which is aim to achieve the following objectives: datasetobjects in the same cluster as similar as possible, and as different as possible betweendifferent clusters. Mount of technology for clustering has been proposed, while the currentSociety Topic is straddling, how to apply clustering techniques to other disciplines hasbecome a hot research, with multi-scale science developing, achieving multi-scale clusteringis becoming increasingly important.Multi-scale clustering has been well researched in the several years. Researcher Sunfirstly summarized multi-scale mining in three ways: convert data into multiple scales first,and mining on all the scales later; mining data with the help of parameter which controls scalebeing mined; mining data first, and convert mining result into multiple scales later. The twoways of multi-scale mining is encountered with a fatal situation, which is mining processmust be applied on every scale. While the3rdway is seldom researched, this paper proposes anew way to get rid of the problem encountered in the first two ways under the idea of3rdway.Works of this paper is as follows:Propose a method to present scale as vector. Scale exists everywhere, no matter what thetype of data is or the database is. The expression of scale is different with different type ofdata, brings the problem of inconvenience of comparison between scales and scaleconversion. With the help of vector, these processes of scale comparison and conversion canbe applied easily.Propose a weighed vector-based multi-scale clustering algorithm to achieve propose ofclustering at different scales. The basic idea of this algorithm is the3rdway of multi-scalemining. First, this algorithm selects a basic scale, and applies clustering algorithm on thebasic scale to maintain cluster knowledge; secondly convert the clustering knowledge to otherscale which is interest for users with the help of scale convert. The proposed algorithm is applied to the analysis of H province floating population tofurther demonstrate the feasibility and effectiveness of the proposed algorithm. Experimentsshow that the algorithm is feasible and effective, and the clustering results can providescientific basis for decision-makers guidance related fields.