Affiliations: LabSTIC, Université de Guelma, Guelma, Algérie | CResTIC, Université de Reims, Reims Cedex, France | LIP6, 4, Place Jussieu, Paris, France
Note: [] Corresponding author. M. Nemissi, LabSTIC, Université de Guelma, B.P. 401. 24000, Algérie. E-mail: [email protected]
Abstract: This paper introduces a neuro-fuzzy framework for handling multi-class classification problems. Instead of decomposing such problems into simple sub-problems and solving each part using a different classifier, the proposed system decomposes and implements the entire problem automatically in the same framework. The decomposition is performed using the most commonly used methods for dividing multi-class classification problems: OAA (one-against-all) and OAO (one-against-one). Consequently, two models are introduced: OAA and OAO based neuro-fuzzy classifiers. The design of the proposed models is based on the implementation of each sub-problem using a set of weights. The learning is performed by adjusting every set independently, and without adjusting the parameters of membership functions. This considerably simplifies the classification and learning tasks. After the learning stage, the proposed systems act as a single-module classifier for recognizing new examples.
Keywords: Pattern recognition, machine learning, multi-class classification, neuro-fuzzy systems
