Differential evolution trained radial basis function network: application to bankruptcy prediction in banks

In this paper, we propose differential evolution (DE) to train the supervised part of the radial basis function (RBF) network in the soft computing paradigm. Here the unsupervised part of the RBF is taken care of by K-means clustering. The new network is named as differential evolution trained radial basis function (DERBF) network. The efficacy of DERBF is tested on bank bankruptcy datasets viz. Spanish banks, Turkish banks, US banks and UK banks as well as benchmark datasets such as iris, wine and Wisconsin breast cancer. The performance of DERBF is compared with that of differential evolution trained wavelet neural networks (DEWNN) (Chauhan et al., 2009), threshold accepting trained wavelet neural network (TAWNN) (Vinaykumar et al., 2008) and wavelet neural network with respect to the criterion area under receiver operating characteristic curve. The results showed that DERBF is very good at generalisation in the ten-fold cross validation for all datasets.