Reproduce “Training Gradient Boosted Decision Trees on Tabular Data Containing Label Noise for Classification Tasks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Anita Eisenbürger, Daniel Otten, Anselm Hudde, Frank Hopfgartner
Researchers adapted and developed methods to detect and correct label noise in Gradient Boosted Decision Trees (GBDTs) on tabular data, achieving over 99% noise detection accuracy on the Adult dataset and showing that while training error is reduced, improvements in classification accuracy vary by dataset.
Key insights
- GBDTs possess a natural robustness to label noise, particularly symmetric noise, highlighting early stopping as a critical defense mechanism.
- The effectiveness of noise detection and correction methods is highly dependent on dataset characteristics, noise type, and noise rate, indicating no universal best approach.
- While explicit noise correction significantly reduces training error, this improvement does not always translate to higher classification accuracy on clean test sets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.