Tree-based methods are convenient and powerful machine learning tools thatcan be seen as alternatives to classical regression and prediction models suchas generalized linear models, see for example. The most standard proceduresare designed to estimate the expectation of a random variable, that is, when itcomes to risk, a central scenario (or a best estimate using the Solvency IIterminology). In this work, we propose an extension of these tree methods tothe study of extreme events, which are of particular interest when it comes toinvestigate the tail of the distribution and design reinsurance policies. Wepropose a detailed description of our adaptation of decision trees and supportthe methodology with new consistency results on these topics. We illustratethe performance of the procedure on simulated data and on a cyberdata base.