Attribute Reduction In Decision Table Based On Granular Computing

In recent years, with a wide range of information processing technology applications and the prevalence of electronic facilities in industry, there is a huge amount of data information. It's very critical for people to discover how to obtain valuable information that is applied to production in daily life. Therefore, a new research field of analyzing data and extracting intelligence information come into being and develope quickly, that is called 'Knowledge Discovery'. The 'Data Mining' has become one of the main research topics of the knowledge discovery.Attribute reduction in data mining or data analysis has an important significance. An information system or decision table may have many reductions, but reduction results will directly affect the data analysis and the rules of the model. People hope to identify information systems or decision table of the minimal reduction, but the minimal reduction solution has been proved to be an NP problem.By studying several main attribute reduction algorithms, most algorithms start from the calculation of the core attribute and expand the attribute set gradually according to different important degrees of attribute. There are different definitions of attribute reduction algorithm, mainly based on the Skoworn A distinction matrix attribute reduction algorithm, based on the importance of attribute reduction algorithm, based on information entropy reduction algorithm and other methods. In this paper, we give a comparative analysis of the current several main differences attribute reduction algorithm and do a further study on how to realize the information systems and decision table reduction algorithm based on Rough Sets and Granular Computing theories. The main achievements of this dissertation include:1. A new knowledge measurement is presented. Based on the viewpoint that attributes have different discernible ability in rough set theory; the concept of knowledge relative distribution is proposed firstly. Its distribution function is based on the intuitionistic diversification of distribution among different knowledge granules and its purpose is that we can observe the situation of knowledge distribution change between different attribute sets. We analyze rationality of the definition, and give some correlative properties. Moreover, based on the concept of knowledge relative distribution, united relative distribution is defined in order to obtain more certain rules after reduction.2. Two different attribute reduction algorithms in information system are proposed. The first algorithm redefines attribute importance according to the relative distribution and takes the new attribute importance as heuristic information and designs a heuristic reduction algorithm; the other algorithm regards united relative distribution as attribute importance and a heuristic reduction algorithm is presented. From the example, the algorithms superiority has been compared and the algorithm characteristic has been analyzed.3. The experimental results show that the algorithm is efficiency and feasibility, and compare advantages and disadvantages of the algorithms. At last, an experimental system has performed on the real data.Finally, the work of this dissertation is summarized and theoretical significance and potential applied value of the research are explained, some deficiencies and the prospective of future research is discussed.