A nodes reduction procedure for RBFNDDA through histogram

This paper presents a two-stage learning algorithm to reduce the hidden nodes of a radial basis function network (RBFN). The first stage involves the construction of an RBFN using the dynamic decay adjustment (DDA) and the second stage involves the use of a modified histogram algorithm (HIST) to reduce hidden neurons. DDA enables the RBFN to perform constructive learning without pre-defining the number of hidden nodes. The learning process of DDA is fast but it tends to generate a large network architecture as a result of its greedy insertion behavior. Therefore, an RBFNDDA-HIST is proposed to reduce the nodes. The proposed RBFNDDA-HIST is tested with three benchmark medical datasets. The experimental results show that the accuracy of the RBFNDDA-HIST is compatible with to that of RBFNDDA but with less number of nodes. This proposed network is favorable in a real environment because the computation cost can be reduced.