Studies On Updating Approaches Of Dominance Rough Set Model Under Dynamic Environments

Rough sets theory is one of effective mathematic tools of dealing with uncertainty and incompleteness data, and it is has been successfully applied in many fields of science and engineering. Traditional rough set theory is based on equivalence relations and can only deal with discrete symbolic valued attributes.In real applications, there are some attributes which have preference orders, such as the attribute “quality” whose values may be “good” or “bad”.. Traditional rough set cannot deal with such kind of data. Greco et al. therefore proposed dominance-based rough sets approach(DRSA) by substitution of the equivalence relation using a dominance relation. Dominance rough sets model is then developed to handle decision problems with preference-ordered data. On the other hand, noticed that the collected real data is often updated from time to time, such as the deletion of invalid objects, revision some attribute values, and insert some new objects, etc. To summarize, these changes mainly occur in the object set and attribute set. In such situations, traditional dominance rough set approach is very time-consuming when it recalculates approximation sets to further update the knowledge base. Therefore, how to efficiently update approximation sets becomes an important study in rough set models.In this thesis, we consider several different types of dynamic environments based on dominance relation rough set model, i.e., add or delete one object in the object set; and other four combined changes of object set and attribute set. In these different conditions, we have given the corresponding updating principles of approximation sets and developed incremental algorithms. The detailed proofs of updating principles are given, and the experiments are carried out with UCI data according to the proposed algorithms. The experimental results show that the updating method proposed in this paper can effectively improve the computational efficiency, and we also discuss the effect of data size on the experimental results.