EUVIMEDEUVIMED entdecken
Crossref

Uncertainty-quantified artificial neural network reanalysis of malachite green adsorption onto sheep manure biochar: SHAP-based feature importance and identification of candidate operating conditions

Mehmet Fatih Dilekoğlu

Environmental Engineering Research · 13.07.2026 · 32 (2) · 260275-0

DOI: 10.4491/eer.2026.275 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Environmental Engineering Research, online erschienen am 13.07.2026. Verfasst von Mehmet Fatih Dilekoğlu. Die Arbeit nennt 13 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Machine-learning surrogates, particularly artificial neural networks (ANNs), are widely applied to adsorption datasets but are often evaluated only by apparent fit, masking overfitting on small samples. This study presents an uncertainty-quantified reanalysis of a previously published 24-point dataset on malachite green (MG) adsorption onto sheep manure biochar (SMB). A 4-10-10-1 ANN (tanh, L-BFGS) predicting equilibrium uptake from pH, dose, temperature and initial concentration C₀ was probed by strictly leak-free repeated 10×5-fold cross-validation, nested cross-validation, bootstrap resampling, and baselines on identical folds. The apparent fit (R² = 0.9948) contrasted sharply with the leak-free repeated-CV (0.751 ± 0.265) and bias-free nested-CV (0.788 ± 0.187) estimates, revealing split sensitivity from the OFAT design. The ANN outperformed every baseline; notably, second-order RSM (−2.88), Ridge-RSM (−1.42) and GAM (−2.66) all gave negative CV R², confirming the ANN complexity is warranted. Bootstrap SHAP identified C₀ (61.2 %, 95 % CI 35.5–78.3) and dose (29.0 %) as dominant, with C₀ top-ranked in 93 % of resamples. Differential evolution returned a candidate region (pH 4.0, dose 1.0 mg, 37.2 °C, C₀ 10 mg L⁻¹; predicted qe 265.6 mg g⁻¹) at distance 0.315 from the nearest point, confirming interpolation, reported as a testable hypothesis.

Zitieren

APA

Dilekoğlu, M. F. (2026). Uncertainty-quantified artificial neural network reanalysis of malachite green adsorption onto sheep manure biochar: SHAP-based feature importance and identification of candidate operating conditions. Environmental Engineering Research, 32(2), 260275-0. https://doi.org/10.4491/eer.2026.275

Vancouver

Dilekoğlu MF. Uncertainty-quantified artificial neural network reanalysis of malachite green adsorption onto sheep manure biochar: SHAP-based feature importance and identification of candidate operating conditions. Environmental Engineering Research. 2026;32(2):260275-0. doi:10.4491/eer.2026.275

BibTeX
@article{Dilekolu2026,
  author = {Dilekoğlu, Mehmet Fatih},
  title = {Uncertainty-quantified artificial neural network reanalysis of malachite green adsorption onto sheep manure biochar: SHAP-based feature importance and identification of candidate operating conditions},
  journal = {Environmental Engineering Research},
  year = {2026},
  volume = {32},
  number = {2},
  pages = {260275--0},
  doi = {10.4491/eer.2026.275}
}
Kostenlos starten →