Details zur Publikation

Kategorie Textpublikation
Referenztyp Buchkapitel
Titel (primär) Neural networks - training and predicting for biological activities of chemicals
Titel (sekundär) Software-Entwicklung in der Chemie 8 = Software Development in Chemistry 8: Proceedings of the 8th Workshop "Computer in Chemistry" Seeheim - Jugenheim/Darmstadt, November 17-19, 1993
Autor Müller, E.; Schüürmann, G.
Herausgeber Jochum, C.
Erscheinungsjahr 1994
Department OEC; COE
Seite von 281
Seite bis 291
Sprache englisch
Abstract Data sets taken from literature are used to analyze particular aspects of the performance of backpropagation neural networks. For network models of qualitative structure-activity relationships it is shown that the weight initialization for the training phase, the network architecture and the descriptor selection may have significant impacts on the final prediction power of the derived model. It is stressed that the prediction capability is the relevant criterion to evaluate the model performance of neural networks, which is not necessarily related to the training success. The results lead to recommendations for the derivation of neural network models with a sufficient capability for prediction.
dauerhafte UFZ-Verlinkung https://www.ufz.de/index.php?en=20939&ufzPublicationIdentifier=26987
Müller, E., Schüürmann, G. (1994):
Neural networks - training and predicting for biological activities of chemicals
In: Jochum, C. (ed.)
Software-Entwicklung in der Chemie 8 = Software Development in Chemistry 8: Proceedings of the 8th Workshop "Computer in Chemistry" Seeheim - Jugenheim/Darmstadt, November 17-19, 1993
Gesellschaft Deutscher Chemiker (GDCh), Frankfurt/Main, p. 281 - 291