EP-based robust weighting scheme for fuzzy SVMS
Shaoyi Zhang, Kotagiri Ramamohanarao, James C. Bezdek · 2010
Support vector machine (SVM) classifiers represent one of the most powerful and promising tools for solving classification problems. In the past decade SVMs have been shown to have excellent performance in the field of data mining. The standard SVM classifier treats all instances equally. However, in many applications we have different levels of confidence in different instances that belong to a particular class. Fuzzy SVMs have been used to recognize the importance of each training instance. Although these schemes are called fuzzy SVMs, they are basically trained by weighted training instances. In this paper we propose a new robust weighting scheme for the class memberships for fuzzy SVM classifier. The weighting scheme is a sophisticated and effective method for weighting the training instances which makes use of highly discriminating patterns called emerging patterns (EPs). Our experiments show that this new weighting method has excellent performance and noise tolerance compared to the weighting scheme previously proposed.