Interpretable Machine Learning for One-Part Fly-Ash/Slag Geopolymer Strength Prediction: Toward Multifunctional Binder Design

Chemistry-informed data-driven models for predicting strength of ambient-cured slag and fly ash geopolymer concrete

Geopolymer concrete (GPC) is a sustainable alternative to Portland cement concrete; however, complex geopolymerization mechanisms and nonlinear strength development under ambient curing make mixture optimization challenging. This study develops a chemistry-informed data-driven framework to predict the 28-day compressive strength of ambient-cured slag/fly ash–based GPC. A dataset of 151 mixtures was compiled incorporating eight input parameters, including key precursor oxide ratios (SiO₂/CaO, SiO₂/Al₂O₃, and CaO/Al₂O₃), which are rarely considered in existing predictive models. Artificial Neural Network (ANN) and Gene Expression Programming (GEP) models were developed and compared. The optimal ANN model (8–2–2–1 architecture) achieved superior predictive accuracy (R² = 0.93, MAE = 2.82), while the GEP model (R² = 0.77, MAE = 5.55) produced an explicit mathematical equation suitable for practical applications. Model reliability was verified experimentally using four new mix designs. Sensitivity analysis identified the SiO₂/CaO ratio as the most influential parameter governing strength development in ambient-cured GPC.

Leakage-Controlled and Applicability-Domain-Aware Machine Learning Optimization of Fly Ash-GGBS Geopolymer Concrete

Fly ash-ground granulated blast-furnace slag (GGBS) geopolymer concrete is a promising low-clinker binder technology, yet its compressive strength is governed by coupled precursor chemistry, activator dosage, liquid-solid balance, aggregate proportioning, curing history, and testing age. These couplings make purely empirical mixture design expensive and make conventional random-split machine-learning validation prone to optimistic conclusions when repeated or compositionally related mixtures occur in compiled databases. This study presents a leakage-controlled, physics-guided, explainable, and applicability-domain-aware workflow for compressive-strength prediction and multi-objective mixture screening of fly ash-GGBS geopolymer concrete. A restarted database containing 548 records, 13 raw input variables, and compressive strength was audited and expanded using 33 physics-guided descriptors representing alkali activation, calcium–silicate balance, liquid-solid proportioning, curing intensity, age transformations, and nonlinear interaction terms. Exact duplicate records were removed before modeling, and a composition-level grouped split was used to evaluate generalization to unseen mixture families, giving 444 training rows, 98 held-out test rows, 124 training groups, 32 test groups, and zero train–test group overlap. Raw-feature SVR, XGBoost, and CatBoost models were compared with physics-guided XGBoost and CatBoost variants. Raw-feature CatBoost achieved the strongest held-out global accuracy, with R² = 0.903, RMSE = 5.915 MPa, and MAE = 4.207 MPa, while the physics-guided CatBoost surrogate was retained for mechanism-aware diagnosis and downstream constrained optimization. Regional analysis showed that physics-guided descriptors reduced low-strength-tail RMSE by 10.19% but worsened high-strength-tail RMSE by 16.04%, revealing that global accuracy masked uncertainty in the performance region most relevant to optimization. SHAP analysis linked these tail behaviors to curing-age interactions, Ca/Si balance, alkali–aluminate balance, chemistry-reactivity descriptors, and high-strength extrapolation effects. An NSGA-II optimization using compiled material cost and CO₂ factors then produced 260 Pareto candidates spanning 31.67-92.75 MPa predicted strength, 0.191-0.277 USD/kg binder material cost, 0.349-0.444 kg CO₂/kg binder material emissions, and 15.94-382.43 curing-severity units. Crucially, all optimized candidates were outside the 99% nearest-neighbor applicability-domain threshold. The results demonstrate that high apparent predictive accuracy is insufficient for responsible data-driven geopolymer mixture design; optimization outputs should be filtered through leakage-aware validation, regional reliability diagnosis, explainable feature attribution, and applicability-domain screening before being advanced to laboratory validation.

Performance Prediction and Ratio Design of Coal-Based Solid Waste Cemented Filling Materials Based on Ensemble Learning

Coal-based solid wastes, including coal gangue and fly ash, can be extensively utilised in cemented backfill materials. However, the slump, bleeding rate, and mechanical strength of these materials depend nonlinearly on the mixture composition, particle size, solids concentration, and curing conditions, complicating the multi-performance mixture design. This study developed an ensemble-learning framework for the target-specific performance prediction and empirical-uncertainty-aware inverse design of coal-based solid-waste cemented backfill materials. A literature-derived database containing 720 observations and 11 predictors was established. After the target-specific filtering of missing responses, 214 observations were available for the slump, 284 for the bleeding rate, and 711 for the uniaxial compressive strength (UCS). Support vector regression (SVR), Bagging-SVR, AdaBoost-SVR, and Stacking-SVR were evaluated using 20 repeated random 80:20 holdout partitions to assess the within-database predictive performance. Bagging-SVR achieved the lowest mean inner-cross-validation RMSE for all three responses. Its mean test R2 values were 0.969, 0.871, and 0.965 for the slump, bleeding rate, and UCS, respectively, with corresponding RMSE values of 2.228 cm, 1.206 percentage points, and 1.575 MPa. SHAP analysis showed that the coal-gangue particle size and solids concentration received the largest model attributions for the slump and bleeding-rate predictions, whereas the cement content and curing time received the largest attributions for the UCS prediction. The selected Bagging-SVR models were subsequently coupled with multi-objective differential evolution incorporating empirical prediction bounds, component mass balance, and target-specific five-nearest-neighbour applicability-domain constraints. The selected compromise candidate had a solids concentration of 79.46% and coal-gangue, fly-ash, and cement dry-solid mass fractions of 63.29%, 24.95%, and 11.76%, respectively. Its predicted slump, bleeding rate, and 28 d UCS were 21.19 cm, 1.85%, and 6.36 MPa, respectively. The nominal empirical upper bound of the bleeding rate was 3.83%, and the lower bound of the UCS was 3.74 MPa, both satisfying their prescribed limits. However, the nominal slump interval of 15.70–26.65 cm was not fully contained within the prescribed range of 18–26 cm.

Interpretable Machine Learning for Predicting Unconfined Compressive Strength of Cemented Tailings Backfill: A Data-Driven Approach with Bayesian Optimization and SHAP Analysis

Machine Learning–Driven Optimization of Low-Carbon Concrete Mixes Incorporating Industrial Waste

The cement industry is a major contributor to global carbon dioxide (CO₂) emissions, necessitating the development of sustainable construction materials with reduced environmental impact. This study proposes a novel integrated framework that combines industrial waste-based low-carbon concrete development with machine-learning-driven strength prediction and mix optimization. Low-carbon concrete mixes were produced by partially replacing ordinary Portland cement with fly ash and ground granulated blast furnace slag (GGBS) at various replacement levels. Experimental investigations were conducted to evaluate compressive strength development at different curing ages. A dataset comprising 300 experimental observations was subsequently employed to develop and compare three machine learning models, namely Random Forest (RF), Support Vector Regression (SVR), and Extreme Gradient Boosting (XGBoost), for compressive strength prediction. The results demonstrate that appropriately designed low-carbon concrete mixtures can achieve comparable or superior long-term strength while substantially reducing cement consumption and associated CO₂ emissions. Among the evaluated models, XGBoost exhibited the highest predictive accuracy, indicating its suitability for sustainable concrete mix optimization. The novelty of this study lies in integrating experimental low-carbon concrete design, environmental assessment, and advanced machine learning techniques within a unified framework to enhance structural performance and sustainability simultaneously. The proposed approach provides an efficient pathway to reduce trial-and-error experimentation and accelerate the adoption of eco-friendly concrete in modern construction practices.

Comparative and Interpretable Machine Learning for Fly Ash Geopolymer Strength: Prediction, Validation and Optimization

An interpretable machine learning framework integrating Extreme Gradient Boosting with Shapley Additive Explanations to predict the 28-day compressive strength of fly ash-based geopolymer concrete (FA-GPC) is developed and experimentally validates.