Unbounded Fuzzy Radial Basis Function Neural Network Classifier

Abstract

Shetty, Balaji S. Mahindrakar, Manisha S. Kulkarni, U. V.Area of pattern recognition deals with the recognition of patterns by using different machine learning algorithms without human intervention. Many different data mining algorithms are used in pattern recognition. The selection of an appropriate and precise algorithm is very crucial. An imprecise algorithm may lead to generate a wrong decision. Recognition can be supervised or unsupervised. This paper presents a novel unbounded fuzzy radial basis function neural network (UFRBFNN) classifier model to perform the supervised classification. This classifier is constructed using fuzzy clustering and further clusters are converted into fuzzy hyperboxes. Fuzzy set hyperboxes (FHBs) represent the neurons in the hidden layer. The creation of these FHBs is based on the unbounded spread from inter-class information and intra-class fuzzy membership function. The proposed approach is faster and independent of the tuning parameters. The output is determined by the union operation of the FHBs outputs which are connected to the class nodes in the output layer. Using K-fold cross-validation, the UFRBFNN model is verified by applying 7 different standard datasets from (UCI) machine learning repository and further by comparing results with well-known radial basis function neural network (RBFNN) variants. The analysis of the result shows that the proposed model provides 5–10% improved training accuracy with previous radial basis function classifiers.