Fuzzy rough granular self organizing map

Abstract

A fuzzy rough granular self organizing map (FRGSOM) is proposed for clustering patterns from overlapping regions using competitive learning of the Kohonen's self organizing map. The development strategy of the FRGSOM is mainly based on granular input vector and initial connection weights. The input vector is described in terms of fuzzy granules low, medium or high, and the number of granulation structures depends on the number of classes present in the data. Each structure is developed by a user defined α-value, labeled according to class information, and presented to a decision system. This decision system is used to extract domain knowledge in the form of dependency factors using fuzzy rough sets. These factors are assigned as the initial connection weights of the proposed FRGSOM, and then the network is trained through competitive learning. The effectiveness of the FRGSOM is shown on different real life data sets.