PARENT SESSION
PS7 - Applied Statistics in Toxicological Research
Sunday, 17 November 2002
8:00 AM to 6:30 PM
Exhibit Hall

(P100) Use of Decision Trees in Determining the Mechanism of Aquatic Toxicity.

Kim, Hyunjoong^*,1, Ren, Shijin¹, ¹ Department of Statistics, University of Tennessee, Knoxville, Tennessee, USA

ABSTRACT- Hierarchical classification methods for the determination of the mechanisms of aquatic toxicity are being developed. The approach taken is primarily the statistical modeling of toxicity mechanisms using different molecular and quantum-chemical descriptors. The use of the method of decision trees for the classification of narcotic and reactive toxicity, two broad mechanism designations that correspond to the first level of the hierarchy, was investigated. Hydrophobicity and electrophilicity descriptors were used in the toxicity mechanism classifications. The results of the classifications were compared with the results obtained by more traditional statistical methods such as discriminant analysis and logistic regression. It was found that the method of decision trees yielded a smaller error rate and that its results were more interpretable.

Key words: mechanism, aquatic toxicity, decision trees, statistical methods