| Rough set theory developed in1980s is a new mathematical tool to deal with vagueness and uncertainty of information. It is different from the classic methods that used to deal with uncertain problems, such as probability model, D-S evidence theory, fuzzy reasoning etc. Its advantage is it extracts the valid and potential useful knowledge in data without anything from others.In this paper, firstly, we study on the attribute reduction algorithm. We main do two aspect work. On one hand, we improve on one of the existed algorithms which named attribute reduction algorithm based on discernility matrix and logic operation. We propose a new transformation method from conjunction matrix to disjunction matrix that avoids the manual work or global search and can save the time and space. On the other hand, most traditional methods didn't integrate with database operations. In this paper, we propose some new methods for computing the attribute cores, the significance of attributes and the dependency between attribute subsets based on database operations. Based on these new computing methods, we also present a new heuristic attribute reduction algorithm. It is extremely efficient for the large decision table.Afterward, we study on the value reduction algorithm and propose a new heuristic value reduction algorithm based on discernility matrix which has good mathematic model and makes the value reduction process disjoint with the decision table. Besides, the decision rules are recognized by digging the statistical information that includes accuracy and coverage and their error is proved to be minimal using accuracy as choosing criterion for uncertain rules.After that we study on the ordered decision table and propose a new heuristic attribute reduction algorithm based on dominance matrix, whose time complexity is polynomial. Meanwhile, the efficiency of the new attribute algorithm is proven through the example.Finally, we carry out a system based on the above algorithms. |
PDF Full Text Request