![[2602.22432] LoBoost: Fast Model-Native Local Conformal Prediction for Gradient-Boosted Trees](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.22432] LoBoost: Fast Model-Native Local Conformal Prediction for Gradient-Boosted Trees
Summary
LoBoost introduces a novel method for local conformal prediction in gradient-boosted trees, enhancing uncertainty quantification without retraining or auxiliary models.
Why It Matters
This research addresses the limitations of traditional conformal prediction methods, which often struggle with heteroscedasticity and data efficiency. By improving the calibration of predictions, LoBoost can significantly enhance the reliability of machine learning models in practical applications, making it a valuable contribution to the field of machine learning.
Key Takeaways
- LoBoost leverages the leaf structure of gradient-boosted trees for efficient local conformal prediction.
- It eliminates the need for retraining or auxiliary models, enhancing data efficiency.
- The method shows competitive interval quality and improved test mean squared error (MSE) across various datasets.
- Calibration speed is significantly increased, making it practical for real-time applications.
- LoBoost's approach is particularly beneficial for handling heteroscedastic data.
Statistics > Machine Learning arXiv:2602.22432 (stat) [Submitted on 25 Feb 2026] Title:LoBoost: Fast Model-Native Local Conformal Prediction for Gradient-Boosted Trees Authors:Vagner Santos, Victor Coscrato, Luben Cabezas, Rafael Izbicki, Thiago Ramos View a PDF of the paper titled LoBoost: Fast Model-Native Local Conformal Prediction for Gradient-Boosted Trees, by Vagner Santos and 4 other authors View PDF HTML (experimental) Abstract:Gradient-boosted decision trees are among the strongest off-the-shelf predictors for tabular regression, but point predictions alone do not quantify uncertainty. Conformal prediction provides distribution-free marginal coverage, yet split conformal uses a single global residual quantile and can be poorly adaptive under heteroscedasticity. Methods that improve adaptivity typically fit auxiliary nuisance models or introduce additional data splits/partitions to learn the conformal score, increasing cost and reducing data efficiency. We propose LoBoost, a model-native local conformal method that reuses the fitted ensemble's leaf structure to define multiscale calibration groups. Each input is encoded by its sequence of visited leaves; at resolution level k, we group points by matching prefixes of leaf indices across the first k trees and calibrate residual quantiles within each group. LoBoost requires no retraining, auxiliary models, or extra splitting beyond the standard train/calibration split. Experiments show competitive interval quality, im...
