Intelligent design routes for carbon-sequestration materials: A stacking machine learning-based predictive framework for compressive strength modeling of biochar-modified cementitious composites

Türk K., Katlav M.

JOURNAL OF CLEANER PRODUCTION, cilt.558, ss.1-20, 2026 (SCI-Expanded, Scopus)

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 558
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1016/j.jclepro.2026.148269
  • Dergi Adı: JOURNAL OF CLEANER PRODUCTION
  • Derginin Tarandığı İndeksler: Scopus, Science Citation Index Expanded (SCI-EXPANDED), Chimica, Compendex, INSPEC, Public Affairs Index
  • Sayfa Sayıları: ss.1-20
  • İnönü Üniversitesi Adresli: Evet

Özet

The transition toward low-carbon and environmentally responsible construction materials has amplified the

importance of biochar-modified cementitious composites (BMCC), which provide long-term carbon sequestration

due to the stable retention of biogenic carbon. However, predicting the compressive strength (CS) of BMCC is

challenging because mixture components—especially biochar-related properties—interact in nonlinear and nonmonotonic

ways. Such complex behavior limits the applicability of traditional empirical correlations and illustrates

the importance of advanced artificial intelligence (AI)-based modeling. This paper develops an explainable

AI-guided framework using four base learners and ten stacking machine-learning (ML) models to estimate the CS

of BMCC based on a curated dataset of 482 instances compiled from 26 independent studies. Among the evaluated

model configurations, the SM-8 stacking model, integrating Extreme Gradient Boosting (XGB), Random

Forest (RF), and Extra Trees Regressor (ETR) base learners, achieved the highest predictive accuracy, yielding

superior test-phase performance with R = 0.972, R2 = 0.945, RMSE = 3.90, MAE = 2.68, and MAPE = 7.84%.

Explainability analyses using SHAP and ICE identified age, w/c, cement, SP, and biochar characteristics as the

dominant factors influencing CS, while also highlighting nonlinear relationships and optimum-performance regions

for critical parameters. In addition, cradle-to-gate life-cycle carbon footprint and production cost indicators

were quantified to support sustainability-oriented mixture assessment. To enhance practical accessibility, a userfriendly

tool was developed, enabling real-time predictions, guided parameter ranges, model selection, and

simultaneous reporting of environmental and economic indicators. Overall, the proposed explainable AI

framework enables accurate BMCC mix optimization, supporting sustainable construction and practical engineering

decision-making.