Research On Evidence Fusion Method Based On Rough Set

Rough set theory and Evidence theory are powerful tools for dealing with uncertain problems. Rough set theory deals with uncertain data without prior information. Because of this advantage it has been widely applied in the field of information fusion. Evidence theory uses basic probability assignment function, belief function and plausibility function to handle uncertain problems. And evidence theory has significant superiority in the field of information fusion. Based on the relationship between rough set and evidence theory, the thesis focuses on methods of evidence acquirement and weight of evidence methods.The thesis describes the concepts and properties of rough set and evidence theory. Then it discusses methods of evidence acquirement and weight of evidence methods based on complete decision table in detail. There are deficiencies in the method of evidence acquirement based on conditional basic probability assignments. Aiming at the deficiencies, a novel method of evidence acquirement and combination based on complete decision table is proposed. Firstly, the concept of confidence degree of evidence is given with the relationship between conditional attributes and decision attribute. Conditional basic probability assignments are calculated according to confidence degree of evidence. Secondly, the concept of support degree of evidence is presented. From horizontal and vertical aspects, the attribute significance and support degree of evidence are used to calculate weight of evidence. Finally, decisions are gained by utilizing combinational rule to integrate conditional basic probability assignments. The results show the validity and superiority of this method.Methods of evidence acquirement are based on complete decision tables. For incomplete decision table, however, basic probability assignment is obtained with a high degree of subjectivity. Therefore, two different methods of evidence acquirement based on incomplete decision tables are proposed to solve this problem. One is under compatibility relation; the other is under non-symmetric similarity relation. And theoretical analysis and examples prove the validity of these algorithms.