A Novel Soft Set Approach for Feature Selection

Feature selection is an important preprocess for data mining. The soft set theory is a new mathematical tool to deal with uncertainties. Merging the soft set theory into the feature selection process based on rough sets facilitates the computation with equivalence classes and improves the efficiency. We propose a paired relation soft set model based on equivalence classes of the information system. Then we use it to present the lower approximate set in the form of soft sets and calculate the degree of dependency between relations. Furthermore, we give a new mapping to obtain equivalence classes of indiscernibility relations and propose a feature selection algorithm based on the paired relation soft set model. Compared to the algorithm based NSS, this algorithm shows 18.17% improvement on an average. Meantime, both of the algorithms show a good scalability.