Stepwise Clustered Ensemble (SCE): An R package for interpretable and robust regression in environmental modeling
Kailong Li, Michael R. Martin, Xiuquan Wang (Xander), Farnaz Hosseinpour · Environmental Modelling & Software · 2026
We present Stepwise Clustered Ensemble (SCE), an open-source R package for regression and inference designed to address key limitations of traditional random forest models. SCE replaces impurity splitting with a likelihood-based criterion using Wilks Λ statistic and adds a node-merging step to prevent superfluous partitions. This statistically rigorous construction enhances parsimony and model generalization. The package is fully open-source and designed to integrate seamlessly into reproducible environmental modelling workflows, facilitating transparency and collaboration. The predictors selected by SCE transfer effectively to Random Forest model in the evaluated case studies, improving its performance and offering a practical route to feature selection. Through large-scale environmental applications in hydrology and air-quality modeling, SCE demonstrates strong predictive performance and interpretability while helping reduce overfitting and improve robustness to noisy environmental conditions. By integrating these innovations into an R package, SCE provides researchers with a robust, interpretable, and transferable framework for modeling complex environmental systems.