EUVIMEDEUVIMED entdecken
Leserbrief

Beyond accuracy: Stabilizing feature importance in GWRF/RF for soil heavy metal mapping

Yoshiyasu Takefuji

Environmental Pollution · 2026-02 · 390 · 127447

DOI: 10.1016/j.envpol.2025.127447 ↗

Kurz gefasst

Leserbrief in der Zeitschrift Environmental Pollution, 2026-02. Verfasst von Yoshiyasu Takefuji. Beteiligte Einrichtung: Faculty of Data Science, Musashino University, 3-3-3 Ariake Koto-ku, Tokyo, 135-8181, Japan. Electronic address: takefuji@keio.jp. Hauptthemen laut MEDLINE: Soil Pollutants, Metals, Heavy, Environmental Monitoring, Soil. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 21 Literaturangaben.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Qin et al. (2025) integrate geographically weighted random forest with multi-source environmental covariates to map soil heavy metals, showing that optimal bandwidth selection and fusion of local GWRF with global RF markedly improve predictive accuracy. Using mean decrease in impurity and SHAP, they identify O3, topography, mining proximity, and rainfall as dominant predictors. However, feature importance lacks ground-truth validation, and rankings are model-dependent, risking misleading interpretation despite strong target prediction. Because SHAP explanations inherit supervised-model biases, interpretability cannot be inferred from accuracy. We recommend rank-stability auditing and label-agnostic validation: generate independent rankings (e.g., feature agglomeration, highly variable feature selection, Spearman's ρ), then perform a leave-one-out rank-stability test to assess order robustness. Transparent reporting of ranking instability should accompany predictive metrics.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Soil PollutantsMetals, HeavyEnvironmental MonitoringSoil

Stichwörter der Autor:innen

Feature importance stabilityGWRFRandom forestSHAPSoil heavy metals

Substanzen

Soil PollutantsMetals, HeavySoil

Autor:innen und Institutionen

  • Faculty of Data Science, Musashino University, 3-3-3 Ariake Koto-ku, Tokyo, 135-8181, Japan. Electronic address: takefuji@keio.jp.

Zitieren

APA

Takefuji, Y. (2026). Beyond accuracy: Stabilizing feature importance in GWRF/RF for soil heavy metal mapping. Environmental Pollution, 390, 127447. https://doi.org/10.1016/j.envpol.2025.127447

Vancouver

Takefuji Y. Beyond accuracy: Stabilizing feature importance in GWRF/RF for soil heavy metal mapping. Environmental Pollution. 2026;390:127447. doi:10.1016/j.envpol.2025.127447

BibTeX
@article{Takefuji2026,
  author = {Takefuji, Yoshiyasu},
  title = {Beyond accuracy: Stabilizing feature importance in GWRF/RF for soil heavy metal mapping},
  journal = {Environmental Pollution},
  year = {2026},
  volume = {390},
  pages = {127447},
  doi = {10.1016/j.envpol.2025.127447}
}
Kostenlos starten →