Class-Specific Attribute Reducts Based On Uncertainty Degrees Of Rough Sets

Attribute reducts are the main content of rough sets theory and an effective means of data analysis,and the related research has important value and significance.The traditional classification-based attribute reducts are applicable to the optimization of all decision classes,in response to the actual local optimization needs,class-specific reducts came into being.The uncertainty of rough sets mainly exists in the boundary domain,while the attribute reducts based on the boundary domain mainly stay in the classification-based reducts,and the related class-specific reducts are rare.As a consequence,this dissertation is based on the construction of rough sets uncertainty,compares the classification-based reducts based on uncertainty degrees,class-specific attribute reducts based on uncertainty degrees are proposed,and studies the reducts of horizontal relationships,hierarchical relationships,and expansion transformations.The main research content involves the following three aspects:(1)For the class-specific attribute reducts,the rough sets boundary is used to discuss.The uncertainty degrees of decision classification are decomposed hierarchically,and thus uncertainty degrees of decision class are established to acquire their basic properties(such as granulation monotonicity).Based on this,class-specific attribute reducts based on uncertainty degrees are proposed,and their heuristic algorithm is designed,examples are given to verify the correctness of the relevant properties.Using 4 types of UCI data sets to conduct experiments to prove the granular monotonicity of class-specific based on uncertainty degrees.(2)From the reducts of horizontal connection,achieve the relationship between classspecific reducts based on uncertainty degrees and class-specific reducts based on positive regions and class-specific reducts based on approximate.In the coordination class,the three reducts are equivalent,and the inconsistent class may cause class-specific reducts based on uncertainty degrees to be stronger than and different from the class-specific reducts based on positive regions and class-specific reducts based on approximate.From the reducts of vertical connection,achieve the relationship between class-specific reducts based on uncertainty degrees and classification-based reducts based on uncertainty degrees.Express the relevance of the reducts conditions,the transformation conditions of the reducts,and the derivation of the classification-based attribute reducts to class-specific attribute reducts.(3)Class-specific reducts based on uncertainty degrees are extended to the interval-valued information system.Define class-specific based on uncertainty degrees in the interval-value information system and study the granulation monotonicity.Then construct class-specific reducts based on uncertainty degrees,the corresponding algorithm is provided and verify example.Based on the above results,the newly proposed class-specific attribute reducts exhibit improvements,the reducts are improved through examples,algorithms and experiments.They are beneficial to the class-specific optimization processing and uncertainty application.