Study of continuous ID3 and radial basis function algorithms for the recognition of glass defects

Two neural network algorithms were applied to the recognition of defects found in manufactured glass and compared with the standard backpropagation algorithm. They were the continuous ID3 (CID3) algorithm and radial basis function (RBF) networks. Backpropagation achieved a recognition rate comparable to that of CID3, but required a comparatively long training time. For classification into two categories, the CID3 algorithm required less time to train. RBF networks can be trained in less time than both backpropagation and CID3, but the accuracy is reduced. In terms of architecture complexity, backpropagation requires that the number of layers and nodes be specified, whereas the architecture of a radial basis function network is implied. Similarly, CID3 creates its own network architecture during training