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)/owner(s).

The reli-ability plots in the bottom of the figure show the function fitted with Isotonic Regression. Examining the histograms of predicted values (top row in Figure 1), note that almost all the values predicted by boosted trees lie in the central region with few predictions

Tags:

  With, Good, Learning, Supervised, Ability, Predicting, Probabilities, Lire, R eliability, Predicting good probabilities with supervised learning

Information

Related search queries