The Research On Clustering Algorithms Based-on Rough Set Theory

With the development of database technology and the widely use of database management system, people need to manage with more and more data. However, currently very few tools can be used for the analysis of these data. So many people want to predict the future trend of existing data by advanced analysis. Thus, data mining was proposed. Data mining is novel and active and is one of the most forward lines of database and information decision research. The rough set theory, which was introduced by Pawlak Z. in 1982, is a useful tool to deal with vagueness and uncertainty. With rough set theory, decision or classification rules can be deduced during the process of knowledge reduction, with classify ability not decreased. Rough set theory are often used with rule induction, classification and clustering.This paper firstly introduces the basic concepts and improvements and applications of data mining and rough set theory, and the rough set based data mining methods. After draw comparisons of rough sets and clustering algorithms, it presents a novel method to BIRCH, by introducing the concept of upper and lower approximation into the BIRCH. This method analyses an attribution problem of data objects, when data objects simultaneously belong to multiple clusters. The experimental results show that the improved BIRCH has accuracy improved.Outlier mining is an important research of data mining. After analysis and comparison of commonly used methods of outliers, it makes an outlier detection method based on K-means and agglomerative clustering for data stream. The result of the experiment shows that this algorithm is feasible and effective.