Predicting Good Probabilities With Supervised Learning

IthacaNY14853 AbstractWe showthatmaxi-mummarginmethodssuchasboost edtreesandboostedstumpspushprobabilityma ssawayfrom0 and1 yieldinga Bayes,whichmake unrealis-ticindependenceassumptions,push probabilitiestoward0 ex-perimentwithtwo waysofcorrectingthebiasedprobabilitiespr edictedbysomelearningmeth- muchdatathey ,randomforests, IntroductionInmany applicationsit isimportanttopredictwellcali-bratedproba bilities;goodaccuracy orareaundertheROCcurve :SVMs,neuralnets,decisiontrees,memory-ba sedlearn-ing,baggedtrees,randomforests,b oostedtrees,boostedstumps,naive show howmaximummarginmethodssuchasSVMs,booste dtrees,andboostedstumpstendtopushpredict edprobabilitiesawayfrom0 predictandyieldsa bayeshave theoppositebiasandtendtopushpredictionsc loserto0 , Bonn,Germany, (s)/

Predicting Good Probabilities With Supervised Learning also justified for boosted trees and boosted stumps. Let the output of a learning method be f(x). To get cali-brated probabilities, pass the output through a sigmoid: P(y = 1jf) = 1 1+exp(Af +B) (1) where the parameters A and B are fitted using maximum

Loading..

Tags:

  With, Good, Learning, Supervised, Predicting, Probabilities, Predicting good probabilities with supervised learning

Information

Related search queries