Association Knowledge Mining Based On Rough Sets

Rough Set theory is a new mathematic tool to deal with fuzzy and uncertain knowledge. Its main thought is to educe problemen decision or sorting rules by knowledge reduction at the premise of keeping sorting capability. The most prominent difference of RS to deal with uncertain problem is that it is unnecessary to offer any transcendent information except data of problem's need. Rough set theory believes that knowledge is a kind of classification that human and other species inhere. One advantage is the capacity of its strong classification[史忠植, 2002]. The mining algorithm of traditional association rules is only applicable to mining bool association rules but cannot directly carry on mining quantitative rules.For these advantages of rough set, we exert rough set to association rules mining. There are approximately following processes: pretreatment —dispersing the successive attribute, attribute reduction—containing two processes , the reduction of attribute set and attribute value, rule picking—association mining.The main work of this paper:(1)1 have investigated the applications of knowledge expression theory in Rough Sets Theory, introduced these concept: knowledge quantum, average knowledge quantum, entropy, entropy difference, united entropy (UE) etc, and made united entropy namely the average knowledge quantum of united expression of conditional attribute sets and decision-making attribute sets, into the applications and serveing as estimating standard of dispersing of successive attribute and reducting of attribute in Roughe Sets.(2) I improve a method of algorithm of dispersing successive attribute—increasing class and decreasing class(ICDC), put forward the algorithm of dispersing successive attribute with united entropy difference(UED). ICDC experience two process: first dispart every attribute into two class, judge wether the sustainability of new attribute sets is equal to that of the first attribute sets, if they are same halt the process of increasing class, otherwise process the increasing class for next attribute until the two sustainability is same. And then process the decreasing class, decreasing a class for very attribute, judge wether the new sustainability is equal to that of the first attribute sets and so on. More over the UED based on property of sustainability and attribute dispersing, only process the decreasing class, early make the quivalence of very attribute as original classification, and then decreases a class of the classification by the method of grade clustering, judge wether thenew UED is equal to or greater than that of the first, if it satisfy then process decreasing the next attribute otherwise halt.(3) I introduce the binary system expression of equal class in order to find the equal class of attribute sets which can be found by and operation of the binary system expression of all attribute of the sets, and found support, interesting and accuracy of association rules. In discovering rules I combine support, interesting and accuracy as the threshold technique of finding association rules.(4)1 bring forward the algorithm of equal class of decision-making attribute to solve equal class of the decision-making table; the algorithm of binary system support to solve the support of association rules, by which the interesting and accuracy can be counted; the algorithm of efficiency association rules to solve the efficency association rules.