An introduction to the predictive technique AdaBoost with a comparison to generalized additive models

M. Kawakita, M. Minami, S. Eguchi, C. E. Lennert-Cody

Research output: Contribution to journalArticle

22 Citations (Scopus)

Abstract

The recently developed statistical learning method boosting is introduced for use with fisheries data. Boosting is a predictive technique for classification that has been shown to perform well with problematic data. The use of boosting algorithms AdaBoost and AsymBoost, with decision stumps, are described in detail, and their use is demonstrated with shark bycatch data from the eastern Pacific Ocean tuna purse-seine fishery. In addition, results of AdaBoost are compared to those obtained from generalized additive models (GAM). Compared to the logistic GAM, the prediction performance of AdaBoost was more stable, even with correlated predictors. Standard deviations of the test error were often considerably smaller for AdaBoost than for the logistic GAM. AdaBoost score plots, graphical displays of the contribution of each predictor to the discriminant function, were also more stable than score plots of the logistic GAM, particularly in regions of sparse data. AsymBoost, a variant of AdaBoost developed for binary classification of a skewed response variable, was shown to be effective at reducing the false negative ratio without substantially increasing the overall test error. Boosting shows promise for applications to fisheries data, both as a predictive technique and as a tool for exploratory data analysis.

Original languageEnglish
Pages (from-to)328-343
Number of pages16
JournalFisheries Research
Volume76
Issue number3
DOIs
Publication statusPublished - 2005 Dec 1
Externally publishedYes

Keywords

  • AsymBoost
  • Boosting
  • Classification
  • Decision stump
  • Logistic regression
  • Shark bycatch

ASJC Scopus subject areas

  • Aquatic Science

Fingerprint Dive into the research topics of 'An introduction to the predictive technique AdaBoost with a comparison to generalized additive models'. Together they form a unique fingerprint.

  • Cite this