This paper reports the results from three artificial neural network models. Levenberg-Marquardt (LM), generalized regression neural networks (GRNN) and learning vector quantization (LVQ) are applied to eight classification problems. The size of the datasets is ranged from 270 to 4177 patterns and the target classes are ranged from 2 to 26 classes. 10-fold cross validation is used to demonstrate the error rate of networks. The experiments show that the generalized regression neural networks outperform the other classifiers, where the average training performance is 0.0436, the testing error rate is 0.137 and the classification rate is 0.85. On the contrary, by using Levenberg-Marquardt, the average training performance is 0.092, the testing error rate is 0.169 and the classification rate is 0.56, and by using learning vector quantization the average training performance is 0.078, the testing error rate is 0.363 and the classification rate is 0.68.
Generalized regression neural networks, generalized regression neural networks, (LM) (LVQ), probabilistic (PNN) networks, radial basis function (RBF) network, Boston Housing Data, Image Segmentation, Abalone, Ecoli
