WIFROWAN: Wrapped Imbalanced Fuzzy-Rough Ordered Weighted Average Nearest Neighbor Classification

Abstract

In this paper we propose an ensemble method based on IFROWANN (Imbalanced Fuzzy-Rough Ordered Weighted Average Nearest Neighbor) algorithm to classify problems with imbalanced data. The ensemble generates many classifiers with different weight strategy and indiscernibility fuzzy relations. Classification is carried out selecting one of three strategies: I- to classify the new instance with the algorithm with best AUC in training. II- to average the memberships of the instance to the fuzzy-rough lower and upper approximation of each class given by the classifiers with best AUC. III- to average the memberships of the instance to the fuzzy-rough lower and upper approximation of each class of the all classifiers. Our method is validated by an extensive experimental study, showing statistically better results than 14 other state-of-the-art methods.