zzy-Rough Neural Networks for Vowel Classification

While desi ning radial basis function neural networks for classification, kzzy clustering is often used to position the hidden nodes in the input space. The main assumption of the clustering is that similar inputs produce similar out uts. In other words, it means that any two in ut patterns t o m the same cluster must be from the same cfass. Generalization is possible in the radial basis function neural networks due to this similarity pro erty. In many real life applications, however, two atterns f?om the same cluster belon6 to different classes, and [ence, classification based on mere similarit pro erty is inadequate. This problem arises because the a d a b l e L t u r e s are not sufficient to discriminate the classes. It implies that the fuzzy clusters generated by the input featuses have rough uncdaint This paper proposes a fuzzy-rough set based network Whig loits fuzzy-rough membership functions to r e duce this r s e m . The proposed network is theoretically a powerful cfassifier as it is equivalent to a u n i d appraximator. Moreover its activity is transparent as it can easily be mapped to a Takagi-Sugeno type fuzzy rule base system. The efficacy of the proposed method is studied on a vowel recognition problem.