Research On Hierarchy Model Of Attribute Reduction In Variable Precision Rough Set

As a new Soft Computing Method, rough set model has been used in many domains widely such as data mining and knowledge discovery. But it is sensitive for datasets which have noise data, therefore some potential useful knowledge can’t be mined. By extending standard inclusion relation to majority inclusion relation, variable precision rough set model allows some degree of misclassification in the largely correct classification, and can deal with datasets having noise data better to discover weak dependence relationship.Attribute reduction in variable precision rough set model becomes more complicated because of the introduction of inclusion degree. The reduction anomalies of current reduction model are analyzed in detail and the reason for reduction anomalies is revealed. To eliminate reduction anomalies, this thesis takes lower approximation distribution as measure of classification ability, redefines reduction model based on lower approximation distribution maintained unchanged and presents the corresponding reduction method. So, the problem of reduction anomalies is converted to the representation of the hierarchical model, and reduction anomalies are eliminated gradually with the deepening of the interval reduction hierarchical model.Attribute reduction in variable precision rough set is interval and the combination of condition classes leads to changing of reduction interval with the reduction of condition attributes. This thesis investigates the influence on reduction interval when condition classes merge together and shows the reason why new reduction model based on lower approximation distribution can avoid interval anomalies, Jumping phenomenon and other kinds of anomalies.As the most important attributes, attribute core in variable precision rough set model is called interval core. The thesis defines interval core based on lower approximation distribution, analyzes its property and presents an algorithm to calculate interval core. On the other hand, interval core can be taken as the beginning subset of Heuristic algorithm. So ordered discernibility matrix is constructed and Heuristic algorithm of attribute reduction is brought forward on lower approximation distribution hierarchy.At last, the application on wine Data Set of reduction model based on quality of classification, relative positive region and lower approximation distribution illustrates the relationship among reductions from this three different reduction model and shows correlative theory further.