Neuro-fuzzy knowledge representation for toxicity prediction of organic compounds
Dan Neagu, Emilio Benfenati, Giuseppina C. Gini, Paolo Mazzatorta, Alessandra Roncaglioni · European Conference on Artificial Intelligence · 2002
Models based on neural and neuro-fuzzy structures are developed to represent knowledge about a large data set containing chemical descriptors of organic compounds, commonly used in industrial processes. The neuro-fuzzy models here proposed include both, QSARs and original numerical values. The developed approaches use various techniques to insert knowledge by training, and to map rules in neuro-fuzzy structures. These possibilities are evaluated and we show that the combination of neuro-fuzzy models, and strategies to insert data in the developed connectionist structures, improve over individual models for toxicity prediction.