Clustering Analysis Of Attribute Reduction On Balance Of Profit And Risk

Rough set is an important mathematical tool to process inconsistent, incompleteand inaccurate information, which proposed by Professor Z. Pawlak. It can producedecision or classification rules by knowledge reduction methods in premise of keepingthe equal classification ability, so it has been applied to many fields successfully, suchas knowledge acquisition, decision analysis, intelligent control, pattern recognition,machine learning and data mining. The attribute reduction of decision table is a crucialissue of rough set theory. General speaking, the attribute reduction can be regarded as aprocess to find a minimal attribute set by removing some unrelated or unimportantattributes. How to find a fast and efficient reduction method is necessary.Clustering analysis, also known as unsupervised classification, is one of the mostwidely used unsupervised data analysis method for inner data structure, and it has beenextensively allied in variety fields, such as computer vision, statistics, image processing,medical, biology, social science and psychology. Many clustering analysis methodsneed some related parameters in advance, such as the cluster number, initial center andthe number of iterations. If lack of those knowledge and experiences, those clustermethods usually are infeasible. Therefore, searching a cluster method which needsrarely prior knowledge, lower interrupt and more accurate result is essential.First, the reduction method of attribute significances based on rough set theory andthe positive region reduction method based on decision-theoretic rough set theory arediscussed. Then, a method on balance of profit and risk is given. After that, the relatedclustering method is put forward. At last, this paper proposes a cluster method ofattribute reduction on balance of profit and risk. The main contributions and originalideas included in the dissertation are summarized as follows.1) Aimed at rough set and knowledge reduction, an attribute reduction algorithmbased on attribute importance is proposed, which could be used in lower noiseinformation system.2) In many practical problems, the domain binary relation is not equivalent, so the application of rough set model is limited. This paper provides a minimum cost of BayesDecision and an attribute reduction on decision-theoretic rough set model. Positiveregion preservation attributes reduction and relates algorithm also is defined in thispaper.3) Usually it is taken grant that we gain the maximal profit and take the minimalrisk in industry, agriculture, economic activities and social life, but the expectation isideal and can’t be achieved. It is an important problem that how to balance profit andrisk, and find out practical attribute reduction in decision-making process. By thealgorithms of attribute reduction, we can find the association relationships of hiddenattributes which simplify the data model and improve system model’s simulationaccuracy. This paper builds a decision-theoretic model which can balance profit and risk,namely, which finds an optimal combination of risk in certain level of expected profit,and then provides a heuristic search algorithm of attribute reduction. The algorithmtakes the function of balance profit and risk as the target of heuristic attribute reduction.Case analysis and experiments show that this algorithm can reduce the scales andcomplexity of data model, make it easier to computer simulation of the model system.Therefore, it has strong practicability and economic value.4) Directed against the shortage of traditional clustering, this paper proposes aclustering method of attribute reduction on balance of profit and risk. At first, we get asmall granularity clustering results by the information table analysis algorithmframework. Then, this paper proposes a clustering algorithm of attribute reduction onbalance of profit and risk which can adjust the threshold value to construct a clusteringevaluation function in order to find the solution to optimize the result. At last, caseanalysis and experiments show the algorithm is feasible.In a word, the dissertation proposes some effective method directed against roughset attribute reduction, positive region preservation reduction in decision-theoreticrough set, attribute reduction and clustering analysis of attribute reduction on balance ofprofit&risk, which further enrich attribute reduction and clustering analysis theory.