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

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...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Machine learning Ročník 104; číslo 2-3; s. 359 - 384
Hlavní autoři: Nikolaou, Nikolaos, Edakunni, Narayanan, Kull, Meelis, Flach, Peter, Brown, Gavin
Médium: Journal Article
Jazyk:angličtina
Vydáno: New York Springer US 01.09.2016
Springer Nature B.V
Témata:
Algorithms
Artificial Intelligence
Calibration
Computer Science
Control
Equivalence
Flexibility
Functionally gradient materials
Learning
Machine learning
Mechatronics
Natural Language Processing (NLP)
Probabilistic methods
Probability theory
Robotics
Simulation and Modeling
Classifier calibration
Boosting
Class imbalance
Cost-sensitive
ISSN:0885-6125, 1573-0565
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí: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.
Bibliografie:SourceType-Scholarly Journals-1
ObjectType-Feature-1
content type line 14
ObjectType-Article-1
ObjectType-Feature-2
content type line 23
ISSN:0885-6125
1573-0565
DOI:10.1007/s10994-016-5572-x