Feature Selection and Weighting Method Based on Similarity Rough Set for CBR

Case-based reasoning systems retrieving cases is an n-ary task. Most researches resolve this problem with a similarity function based on KNN rules or some derivatives. But the result of this method is sensitive to those irrelevant or noisy features. Standard rough set has been used in feature reduct and selection in various domains. But the indispensable discretization ruins the objectivity and the usually used post approximation based weighting method costs lots of computing capacity. This paper proposes a feature selection and weighting method based on similarity rough set theory. It avoids discretizing continuous attributes and keeps the objectivity and quality of datasets. Based on the indiscernibility relation, this method reducts and weighs attributes at the same time. It is easy to realize and can generate accurate results