Modified fuzzy hyperline-segment neural network for classification with mixed attribues

Abstract

The fuzzy hyperline segment neural network (FHLSNN) utilizes fuzzy sets as pattern classes in which each fuzzy set is an union of fuzzy set hyperline segments. The fuzzy set hyperline segment is a n-dimensional hyperline segment defined by two end points with a corresponding membership function. In FHLSNN, membership function calculates membership value of the input pattern based on its distance from both the end points of the hyperline segment. But sometimes input pattern is nearer to the hyperline segment but far from its endpoints. To solve this problem, this paper proposes modified fuzzy hyperline segment neural network (MFHLSNN). In MHLSNN membership function is based on minimum of the distance of the input pattern from the midpoint of the hyperline segment and its distance from both the end points. The proposed model is applied to eight different benchmark datasets taken from the UCI machine learning repository. The experimental results of the MFHLSNN are compared with earlier methods like fuzzy min-max neural network, generalized fuzzy min-max neural network and fuzzy hyperline segment neural network. These results show that the MFHLSNN gives improved performance as compared to its earlier methods.