A Rough-fuzzy Hybrid Approach on a Neuro-Fuzzy Classifier for High Dimensional Data
Faculty of Computing, Health and Science
School of Computer and Security Science / Artificial Intelligence and Optimisation Research Centre
A new Rough-Neuro-Fuzzy (RNF) classifier is proposed in this paper for pattern classification scheme on high dimensional data as an extension of the previous work. The rough set theory is utilized to reduce the given knowledge into a compact form and to obtain a minimal set of decision rules. The proposed Rough-Neuro-Fuzzy classifier is constructed based on the structure of ANFIS (Adaptive-Network-Based Fuzzy Inference System), except its connections determined by the reduced data and the generated decision rules obtained by the rough sets-based approach. This provides the compact and minimal number of configurations for the network to adjust itself towards a faster learning. The learning scheme for the proposed approach is adopted from the one in ANFIS. The TS-type fuzzy inference model is employed to perform the decision making process. The proposed system is applied on a number of data sets for pattern classification tasks using 10-fold cross validation. The number of attributes is reduced significantly and the minimal rules are generated effectively by the rough set-based approach in the proposed system. Experimental results showed that results produced by the proposed rough-neuro-fuzzy classifier may be competitive compared to the previous work and the other existing approaches.