Mixture Kernel-Based Fuzzy-Rough Feature Selection

Fuzzy-rough sets are a hybridisation of fuzzy sets, which encapsulate the related but distinct concepts of fuzziness and indiscernibility in the case of rough sets; both of which occur due to uncertainty in information or knowledge. In recent years, the application of fuzzy-rough sets to the task of feature selection in particular, has demonstrated much success by employing measures based upon the dependencies of features to guide the selection process. Although promising, most existing fuzzy-rough feature selection methods perform only at the level of a single kernel similarity function. As a result, the variability among features is essentially ignored. To address this problem, a mixture kernel-based fuzzy-rough feature selection method is proposed in this paper, using a mixture kernel function. The task of feature selection is accomplished by combining the mixture kernel function and fuzzy-rough sets. The comparative experimental results show an improvement in performance for classification accuracy over traditional fuzzy-rough feature selection approaches.