Using stabilized classification trees in the field of radioecology

Titre du congrès :IASC 07 - Statistics for Data Mining, Learning and Knowledge Extraction. Ville du congrès :Aveiro Date du congrès :30/08/2007

Upon occurrence of a nuclear accident, it is necessary to assess the impact of radioactive deposit and transfer through the environment, particularly through agricultural productions. The purpose of this study is to propose a method to identify the various environmental and anthropogenic characteristics which will mainly influence the radioactive contamination level of plants. A methodology, transposable to other non-agricultural fields, has been developed based on a classification-tree method and the use of radioecological transfer models. In particular, to avoid the problems of instability of decision trees and to preserve their structure, a node-level stabilizing procedure is used.