Giráldez, J. IgnacioElkan, CharlesBorrajo Millán, Daniel2010-03-162010-03-161999-03AAAI Spring Symposium on Predictive Toxicology, AAAI Press: Stanford, March 1999, p. 82-85http://hdl.handle.net/10016/7370Proceeding of: AAAI Spring Symposium on Predictive Toxicology, AAAI Press, Stanford, March 1999A wide panoply of machine learning methods is available for application to the Predictive Toxicology Evaluation (PTE) problem. The authors have built four monolithic classification systems based on Tilde, Progol, C4.5 and naive bayesian classification. These systems have been trained using the PTE dataset, and their accuracy has been tested using the unseen PTE1 data set as test set. A Multi Agent Decision System (MADES) has been built using the aforementioned monolithic systems to build classification agents. The MADES was trained and tested with the same data sets used with the monolithic systems. Results show that the accuracy of the MADES improves the accuracies obtained by the monolithic systems. We believe that in most real world domains the combination of several approaches is stronger than the individuals. Introduction The Predictive Toxicology Evaluation (PTE) Challenge (Srinivasan et al. 1997) was devised by the Oxford University Computing Laboratory to test the suitability ...application/pdfeng© AAAIA Distributed Solution to the PTE Problemconference outputopen access