Obtaining Calibrated Probabilities from Boosting

@inproceedings{citeulike:2883972, abstract = {Boosted decision trees typically yield good accuracy, precision, and ROC area. However, because the outputs from boosting are not well calibrated posterior probabilities, boosting yields poor squared error and cross-entropy. We empirically demonstrate why AdaBoost predicts distorted probabilities and examine three calibration methods for correcting this distortion: Platt Scaling, Isotonic Regression, and Logistic Correction. We also experiment with boosting using log-loss instead of the usual...}, author = {Mizil, Niculescu A. and Caruana, R. }, booktitle = {Proc. 21st Conference on Uncertainty in Artificial Intelligence (UAI '05)}, citeulike-article-id = {2883972}, keywords = {boostexter}, posted-at = {2008-06-12 00:15:09}, priority = {2}, publisher = {AUAI Press}, title = {Obtaining Calibrated Probabilities from Boosting}, url = {http://citeseer.ist.psu.edu/742675.html}, year = {2005} }

TY - CONF ID - citeulike:2883972 L3 - citeulike-article-id:2883972 TI - Obtaining Calibrated Probabilities from Boosting T2 - Proc. 21st Conference on Uncertainty in Artificial Intelligence (UAI '05) PB - AUAI Press N2 - Boosted decision trees typically yield good accuracy, precision, and ROC area. However, because the outputs from boosting are not well calibrated posterior probabilities, boosting yields poor squared error and cross-entropy. We empirically demonstrate why AdaBoost predicts distorted probabilities and examine three calibration methods for correcting this distortion: Platt Scaling, Isotonic Regression, and Logistic Correction. We also experiment with boosting using log-loss instead of the usual... KW - boostexter AU - Mizil, Niculescu AU - Caruana, R PY - 2005/// UR - http://citeseer.ist.psu.edu/742675.html ER -