Computer Science
8 years ago

Boosting by resampling outperforms reweighting for imbalanced data classifiers.

Resampling or Reweighting: A Comparison of Boosting Implementations
Chris Seiffert, Taghi M. Khoshgoftaar, Jason Van Hulse, Amri Napolitano

Paper Summary

Related papers

Theme background image
Computer Science
2 years ago

Boosting Algorithms Revolutionize Predictive Modeling, Outperform Gradient Descent

Transfer Learning
Extreme Learning Machines
Compressed Sensing
Theme background image
Engineering
2 years ago

New boosting method outperforms gradient boosting in accuracy and speed

Transfer Learning
Compressed Sensing
Optimization Methods
Theme background image
Computer Science
2 years ago

Newton boosting trumps gradient boosting for better predictive accuracy in data science.

Learning with Noisy Labels
Bayesian Networks
Explainable AI

Boosting can improve classifier performance, especially with imbalanced data. Two boosting methods exist: reweighting and resampling. Resampling can work with any base learner, while reweighting needs learners that handle example weights. In a study comparing the two methods with imbalanced data, resampling was found to perform as well as or better than reweighting. Therefore, resampling is generally preferred over reweighting for boosting.