Research On Data Mining Classification Algorithm Based On Granular Hierarchy

Granular computing, as an emerging research field, provides a conceptual framework for studying many issues in data mining. The concept of granular computing has been defined and studied by many authors from different points of views, using different notions, based on different conceptual models, and in different contexts. Recently, rough set theory has become a popular mathematical framework for granular computing.Classification is one of the main tasks in machine learning, data mining and pattern recognition. Knowledge for classification can be expressed in different forms, such as classification rules, discriminant functions, decision trees and decision Graphs. On the basis of Rough set theory and Granular Computering theory, this dissertation analyzes the Data mining algorithms for the classification.The main contributions of thesis can be summatized as blow:First, some basic concepts of data mining and data mining process were introduced. From the different task of DM, there are several patterns follows: Classification Pattern, Prediction Pattern, Association Rule Pattern, Regression Pattern, Clustering Pattern, Time Series Pattern and so on. Algorithm is very important for any pattern. The algorithm of Classification Pattern includes: Decision tree, Bayes Classification, Support Vector Machine, Artificial Neural Network, Genetic Alogrithm, Rough set theory.The second chaper, the background of rough set and some basic concept of granular computering were introduced. This paper introduces multi-layered granulations concept.The third chaper, the basic concept of attribute reduction and attribute reduction base on garanular computing were introduced. Level construction attribute reduction and the acquisition of decision rules, the key processes of the applying of granular computering, were studied.In the follwing two chapters, a decision tree classification alortithm base on granule hierarchy was introduced. Based on the granular computing model, we provide a formal and more systematic study of granule centered strategies for the induction of classification rules. In a granule hierarchy, each node is labeled by a subset of objects. The arc leading from a larger granule to a smaller granule is labeled by an atomic formula. In addition, the smaller granule is obtained by selecting those objects of the larger granule that satisfy the atomic formula. The family of the smallest granules thus forms a conjunctively definable covering of the universe. Based on the granule hierarchy, rules for each hierarchical level are induced from data. Then, for each given class, rules for all the hierarchical levels are integrated into one rule. The proposed method was evaluated on a UCI data set, the experimental results of which show that this algorithms is more efficiency.