Back-propagation neural networks-recognition vs. prediction capability
Gerrit Schüürmann, Eckhard Muller · Environmental Toxicology and Chemistry · 1994
Abstract Literature data on biodegradation kinetics of organic compounds, together with a descriptor representation, are subjected to a systematic analysis of the performance of back-propagation neural-network models. The results show distinct dependencies on various model parameters, particularly on the number of iteration cycles. The application of leave-n-out procedures leads to general recommendations for a proper evaluation of the recognition and prediction power of this class of nonlinear structure-activity models.