Abstract
In this paper, the fundamental idea of Incremental Granular Models (IGM) introduced by Pedrycz and Kwak (2007) is followed and their comprehensive design framework is developed. In contrast to typical rule-based systems encountered in fuzzy modeling, the underlying principle of IGM is to consider a two-phase development. First, we build a Linear Regression (LR) model which could be treated as a global model. Next, all modeling errors are compensated by a collection of fuzzy rules that capture more localized nonlinearities of the system as a local model. Here we expand local granular models into Radial Basis Function Networks (RBFN) or Adaptive Neuro-Fuzzy Networks (ANFN) with the use of information granules through Context-based Fuzzy C-Means (CFCM) clustering in the design of incremental model. Numerical studies concern two datasets coming from the machine learning repository. The experimental results revealed that both Incremental RBFN (IRBFN) and Incremental ANFN (IANFN) showed good performance in comparison to the previous works.