Fusion of progressive granular neural networks for pattern classification

Evolving granular neural network (eGNN) is a system that classifies the data by adapting its architecture according to the input information. eGNN posses the ability to learn from the data with a single pass. However, eGNN has the disadvantage of linear updation of weights during training stage that may not handle the real-time uncertain and imprecise data appropriately. To address this issue, the present study aims to use multiple numbers of progressive granular neural networks (PGNNs) in the framework of fusion of classification systems. PGNN is a modified version of eGNN, where the learning process is nonlinear and takes the class-sensitive (CS) granulated data unlike eGNN. Collective opinion from the group of PGNNs increases the classifier performance in comparison with individual PGNNs. The proposed model thus takes the advantages of CS granulation, structural adaptability, nonlinear updation of weights and collective opinions from the group of PGNNs. Performance of model is tested on various datasets, and its superiority to similar other methods has been justified. Various performance indices, such as overall accuracy, dispersion score, kappa coefficient, producer’s accuracy, and user’s accuracy, have been used for performance analysis.