Modified Fuzzy Hypersphere Neural Network for Pattern Classification using Supervised Clustering
Deepak Mane, Uday V. Kulkarni · Procedia Computer Science · 2018
Pattern Classification involves formulating a method that maps the input feature variables to output space of binomial class. In this paper a modified fuzzy hypersphere neural network (MFHSNN) is proposed for pattern classification. MFHSNN utilizes fuzzy set hypersphere as a pattern cluster and classes are represented by a combination of fuzzy set hypersphere. The major factors which improve the learning algorithm of MFHSNN are:a new hypersphere created based on using supervised clustering and patterns of each class decided by its modified fuzzy membership function. If the new input pattern is outside of generated hypersphere then the radius is expanded to include the pattern based on expansion criteria. The presented model tested on three benchmark pattern data-sets and its performance is found to be superior than existing models. The obtained results show that the proposed membership function is able to improve MFHSNN for representation of pattern classification.