Title: Addressing Imbalanced Regression Using Hoeffding Trees

Abstract: A variety of practical scenarios involve the continuous generation of data streams suited for regression tasks. Hoeffding trees, along with their derivatives, have maintained a prominent role in this domain due to their proven efficacy, whether deployed independently or utilized as foundational components within larger ensemble frameworks. Recent studies in batch learning have demonstrated that kernel density estimation (KDE) enhances the quality of smoothed predictions in contexts of imbalanced regression [Yang et al., 2021], while hierarchical shrinkage (HS) offers post-hoc regularization for decision trees without altering their underlying architecture [Agarwal et al., 2022]. This study adapts KDE for streaming environments through a telescoping approach and incorporates HS into incremental decision tree models. Evaluations conducted on established online regression benchmarks indicate that KDE yields consistent improvements in initial stream performance, whereas HS delivers only marginal benefits. The code for this work is accessible at: https://github.com/marinaAlchirch/DSFA_2026.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC