Table 1 Empirical review of existing methods.
Reference | Method Used | Model Type | Dataset Size | Reported R² / MSE | Findings | Interpretability Features | Limitations |
|---|---|---|---|---|---|---|---|
Alam, M.F. et al.36 | Nano-silica analysis and prediction | Statistical + Empirical | 200 | R² = 0.80 | Improved mechanical properties in self-compacting concrete | Not Reported | Poor generalizability to non-nano systems |
Islam, N. et al.37 | Deep Learning Techniques | CNN/ANN | 850 | R² ≈ 0.85, MSE ≈ 3.00 | High-performance concrete strength prediction | Limited (Black-box model) | Low interpretability |
Jibril, M.M. et al.3 | Evolutionary Computational Intelligence | Metaheuristic-based ANN | 900 | R² = 0.84, MSE = 3.20 | Good predictive accuracy for high-strength concrete | Not Explicit | Implementation complexity |
Jubori, D.S.A. et al.38 | Machine Learning Technique | Decision Trees / RF | 700 | R² ≈ 0.82 | Reliable blended concrete predictions | Limited | Requires extensive preprocessing |
Kashem, A. et al.39 | Hybrid ML with SHAP | Ensemble + SHAP | 850 | R² = 0.88, MSE = 2.80 | Strong performance with global interpretability | SHAP | Computational overhead |
Kazemifard, S. et al.40 | Maturity Method with NDT | Empirical | 400 | R² = 0.81 | Effective for specific curing age conditions | Not Included | Method-specific limitation |
Khodaparasti, M. et al.41 | Improved Random Forest | Ensemble Tree-Based | 750 | R² ≈ 0.87 | High accuracy on diverse mixes | Not Discussed | Complex model tuning |
Li, E. et al.8 | XGBoost + Squirrel Search | Optimization + ML | 1000 | R² = 0.86, MSE = 3.00 | Effective for sustainable concrete prediction | Partial (Feature ranking) | Algorithmic complexity |
Li, M. et al.42 | ANN-Based Forecast (Temperature) | Feedforward Neural Network | 500 | R² = 0.84 | Reliable for cooling control | Not Relevant | Limited to thermal analysis |
Lu, C43. | LS-SVR (Automated) | Support Vector Regression | 720 | R² = 0.87 | Accurate high-performance concrete prediction | Not Explained | High computational load |
Meng, X44. | AutoML for HPC | AutoML Framework | 1000 | R² = 0.89 | Enhanced performance through model automation | Implicit | Dependent on data quality |
Nigam, M. et al.45 | Random Forest | Tree-Based Ensemble | 780 | R² = 0.88 | Accurate for nano-silica concrete | Low | Poor explainability |
Onyelowe, K.C. et al.46 | GRG-Optimized RSM | Optimization + Statistical | 600 | R² = 0.86 | Design chart for concrete mix optimization | None | Needs heavy experimentation |
Sapkota, S.C. et al.47 | Ensemble ML | Voting Ensemble | 900 | R² ≈ 0.87 | Robust prediction for normal concrete | Not Included | Long training time |
Saxena, A. et al.15 | Regression Analysis | Statistical | 800 | R² = 0.88, MSE = 2.90 | Strong correlation with sustainable concrete properties | Not Included | Limited to linear patterns |
Shubham, K. et al.48 | Deep Neural Networks | DNN | 950 | R² = 0.87, MSE = 2.95 | High accuracy for industrial waste-incorporated concrete | None | High data requirements |
Tabrizikahou, A. et al.49 | Soft Computing Techniques | Hybrid ANN + GA | 650 | R² ≈ 0.85 | Good predictions for shear strength | Limited | Complex model structure |
Xue, X. et al.50 | Soft Computing | Neuro-Fuzzy Systems | 700 | R² = 0.86 | Accurate bond strength prediction for FRP reinforced concrete | Not Discussed | Heavy feature engineering |
Yao, G. et al.51 | Strut-and-Tie Model | Mechanistic | 300 | R² = 0.82 | Reliable beam strength prediction | None | Limited applicability |
Yu, L52. | Hybrid Regression Framework | Hybrid Regression | 700 | R² = 0.85 | Effective prediction for RCA concrete | Low | Interpretability challenges |
Zhou, J. et al.53 | NDT with AI | Signal Processing + AI | 400 | N/A | Accurate rebar depth and location estimation | Not Relevant | NDT-specific setup |
Chen, C. et al.54 | Hybrid DNN for Tunneling Machines | DNN with Task-Specific Blocks | 800 | R² = 0.90 | Accurate force prediction in tunnel boring | None | High computation |
Diksha et al.55 | Alccofine-Based Geopolymer Concrete | Empirical + ML | 650 | R² ≈ 0.86 | Good performance in geopolymer-based concretes | Not Included | Requires material-specific tuning |
Imran, M. et al.56 | ML Algorithms for HPC | Multiple ML models | 1000 | R² = 0.88 | Improved accuracy over empirical formulas | Implicit | High data dependency |
Proposed Model (This Study) | Hybrid XGBoost + DNN + AutoML + MTL | Ensemble + Deep Learning | 1030 | R² = 0.91, MSE = 2.45 | Accurate, interpretable, robust; SHAP & LIME used effectively | SHAP + LIME | Limited to datasets with detailed input params |
