Optimal sub-reducts in the dynamic environment

Test cost is what we pay for collecting a data item of an object. It is one of the most important factors in cost-sensitive learning. In many classification applications, there is a test cost constraint due to limited money, time, or other types of resources. Hence we need to construct test sets meeting the constraint, and at the same time, preserving the information of the decision system to the highest degree. This problem has been addressed recently under the framework of cost-sensitive rough sets. In this paper, we consider the dynamic environment where both test costs and the constraint change over time. Whenever such changes take place, optimal test sets should be recomputed. We propose a two-stage approach to this environment. The first stage computes all sub-reducts using a backtrack algorithm. The algorithm is time consuming, however it executes only once. The second stage selects optimal sub-reducts according to test costs and the constraint. It is very efficient and executes every time changes are made. We compare the performance of the new approach with an existing one. Experimental results on four UCI (University of California - Irvine) datasets indicate that the new one is more efficient in the dynamic environment.