Extend DICS to regression settings

Extend Data-Informed Centroid Splitting (DICS), currently designed to generate data-informed candidate splits for classification trees, random forests, and gradient-boosting models, to regression settings in order to improve the efficiency of regression-tree methods.

Background

Data-Informed Centroid Splitting (DICS) accelerates tree-based classification by using clustering-derived centroid boundaries to construct a compact set of candidate feature-threshold splits. The theoretical analysis and experiments in the paper concern classification tasks and demonstrate reduced training costs with comparable predictive performance to exhaustive split search.

The paper explicitly identifies the classification-only scope of the current approach as its main limitation and states that extending data-informed priors to regression is an important future direction. Because no regression formulation or results are provided, adapting DICS to regression remains unresolved.

References

Currently, the proposed approach focuses on accelerating classification tasks, which serves as the main limitation. An important direction for future work is to extend this framework by incorporating data-informed priors to similarly improve efficiency in regression settings.

— DICS: Data-Informed Centroid Splitting for Decision Tree Classifiers  (2608.20258 - Mazumder et al., 20 Aug 2026) in Section Conclusion and Future Directions, Section 4