Skip to content

Cost-sensitive boosting algorithms: Do we really need them?

Research output: Contribution to journalArticle

Standard

Cost-sensitive boosting algorithms : Do we really need them? / Nikolaou, Nikolaos; Edakunni, Narayanan; Kull, Meelis; Flach, Peter; Brown, Gavin.

In: Machine Learning, Vol. 104, No. 2, 01.09.2016, p. 359-384.

Research output: Contribution to journalArticle

Harvard

Nikolaou, N, Edakunni, N, Kull, M, Flach, P & Brown, G 2016, 'Cost-sensitive boosting algorithms: Do we really need them?', Machine Learning, vol. 104, no. 2, pp. 359-384. https://doi.org/10.1007/s10994-016-5572-x

APA

Nikolaou, N., Edakunni, N., Kull, M., Flach, P., & Brown, G. (2016). Cost-sensitive boosting algorithms: Do we really need them? Machine Learning, 104(2), 359-384. https://doi.org/10.1007/s10994-016-5572-x

Vancouver

Nikolaou N, Edakunni N, Kull M, Flach P, Brown G. Cost-sensitive boosting algorithms: Do we really need them? Machine Learning. 2016 Sep 1;104(2):359-384. https://doi.org/10.1007/s10994-016-5572-x

Author

Nikolaou, Nikolaos ; Edakunni, Narayanan ; Kull, Meelis ; Flach, Peter ; Brown, Gavin. / Cost-sensitive boosting algorithms : Do we really need them?. In: Machine Learning. 2016 ; Vol. 104, No. 2. pp. 359-384.

Bibtex

@article{6a988392a1db41079768f8d0d9f52c29,
title = "Cost-sensitive boosting algorithms: Do we really need them?",
abstract = "We provide a unifying perspective for two decades of work on cost-sensitive Boosting algorithms. When analyzing the literature 1997–2016, we find 15 distinct cost-sensitive variants of the original algorithm; each of these has its own motivation and claims to superiority—so who should we believe? In this work we critique the Boosting literature using four theoretical frameworks: Bayesian decision theory, the functional gradient descent view, margin theory, and probabilistic modelling. Our finding is that only three algorithms are fully supported—and the probabilistic model view suggests that all require their outputs to be calibrated for best performance. Experiments on 18 datasets across 21 degrees of imbalance support the hypothesis—showing that once calibrated, they perform equivalently, and outperform all others. Our final recommendation—based on simplicity, flexibility and performance—is to use the original Adaboost algorithm with a shifted decision threshold and calibrated probability estimates.",
keywords = "Boosting, Class imbalance, Classifier calibration, Cost-sensitive",
author = "Nikolaos Nikolaou and Narayanan Edakunni and Meelis Kull and Peter Flach and Gavin Brown",
year = "2016",
month = "9",
day = "1",
doi = "10.1007/s10994-016-5572-x",
language = "English",
volume = "104",
pages = "359--384",
journal = "Machine Learning",
issn = "0885-6125",
publisher = "Springer Verlag",
number = "2",

}

RIS - suitable for import to EndNote

TY - JOUR

T1 - Cost-sensitive boosting algorithms

T2 - Do we really need them?

AU - Nikolaou, Nikolaos

AU - Edakunni, Narayanan

AU - Kull, Meelis

AU - Flach, Peter

AU - Brown, Gavin

PY - 2016/9/1

Y1 - 2016/9/1

N2 - We provide a unifying perspective for two decades of work on cost-sensitive Boosting algorithms. When analyzing the literature 1997–2016, we find 15 distinct cost-sensitive variants of the original algorithm; each of these has its own motivation and claims to superiority—so who should we believe? In this work we critique the Boosting literature using four theoretical frameworks: Bayesian decision theory, the functional gradient descent view, margin theory, and probabilistic modelling. Our finding is that only three algorithms are fully supported—and the probabilistic model view suggests that all require their outputs to be calibrated for best performance. Experiments on 18 datasets across 21 degrees of imbalance support the hypothesis—showing that once calibrated, they perform equivalently, and outperform all others. Our final recommendation—based on simplicity, flexibility and performance—is to use the original Adaboost algorithm with a shifted decision threshold and calibrated probability estimates.

AB - We provide a unifying perspective for two decades of work on cost-sensitive Boosting algorithms. When analyzing the literature 1997–2016, we find 15 distinct cost-sensitive variants of the original algorithm; each of these has its own motivation and claims to superiority—so who should we believe? In this work we critique the Boosting literature using four theoretical frameworks: Bayesian decision theory, the functional gradient descent view, margin theory, and probabilistic modelling. Our finding is that only three algorithms are fully supported—and the probabilistic model view suggests that all require their outputs to be calibrated for best performance. Experiments on 18 datasets across 21 degrees of imbalance support the hypothesis—showing that once calibrated, they perform equivalently, and outperform all others. Our final recommendation—based on simplicity, flexibility and performance—is to use the original Adaboost algorithm with a shifted decision threshold and calibrated probability estimates.

KW - Boosting

KW - Class imbalance

KW - Classifier calibration

KW - Cost-sensitive

UR - http://www.scopus.com/inward/record.url?scp=84982845011&partnerID=8YFLogxK

U2 - 10.1007/s10994-016-5572-x

DO - 10.1007/s10994-016-5572-x

M3 - Article

AN - SCOPUS:84982845011

VL - 104

SP - 359

EP - 384

JO - Machine Learning

JF - Machine Learning

SN - 0885-6125

IS - 2

ER -