Table 1 Summary of different ML models used for predicting concrete strength incorporated with FS and CBA.
References | Model used | Input variables | Output variables | Waste materials used | Dataset size | Key findings | Gaps identified |
|---|---|---|---|---|---|---|---|
Neural network & support vector regressor | Slump, WFS%, w/c, cement, aggregate content, curing age | Compressive, split, and flexural strength | FS | 137 | Neural network outperformed support vector in predicting compressive, tensile, and flexural strength; curing time was identified as the most influential parameter. High R2 and low RMSE were achieved for all strength properties | Single-waste material (FS only); no ensemble learning methods used; no dual-waste interaction (e.g., FS + CBA); limited generalizability due to no external validation on a heterogeneous dataset | |
Hybrid neural network | w/c, cement, cement class, WFS%, aggregate content, curing age | Compressive strength | FS | 340 | Optimised neural network gave co-relation = 0.9971, lowest error range (± 1.5 MPa); w/c ratio and cement content were most influential | Only FS; no ensemble models; no feature interpretability; no dual-waste modeling; compressive strength only | |
Artificial neural network | FS/cement, w/c, fine to total aggregate ratio, coarse aggregate to cement ratio, FS/fine aggregate, curing age | Compressive, split tensile, and flexural strength | FS | 397 | The models accurately predicted multiple concrete properties, better than multi-layer regressor model | Only FS used; no ensemble/hybrid models; no feature importance; interpretability and external validation not discussed | |
Hybrid gradient boosting | Cement, FS%, aggregate content, w/c, curing age | Compressive strength | FS | 430 | The optimised technique outperformed in prediction strength | Only FS used; no dual-waste modeling; lacks interpretability (feature importance); limited benchmarking vs. ensemble models | |
Artificial neural network | Cement, aggregate content, fly ash%, CBA%, w/c | Compressive strength | CBA, fly ash | 111 | ANN predicted compressive strengths at multiple ages using literature | Focused only on CBA and Fly Ash; no ensemble or hybrid ML; no dual-waste approach; limited interpretability beyond input ranking | |
Hybrid support vector regressor | Cement, aggregate content, CBA%, w/c, curing age | Compressive strength | CBA | 156 | Hybrid model outperformed; showed sand, curing age, and cement most influential; 25–50% CBA improved compressive strength at 90 days | Only CBA considered; no dual-waste or ensemble ML; no benchmarking vs. other models; dataset size not detailed; limited external validation | |
Artificial neural network | Fly Ash, CBA, water to binder ratio, aggregate content | Compressive strength | CBA, fly ash | 46 | The model achieved best performance for compressive strength prediction of geopolymer concrete | No ensemble/hybrid ML; interpretability not addressed; | |
Random forest & gradient boosting | Density, w/c ratio, fly ash%, CBA% | Compressive strength | CBA, fly ash | 93 | compressive strength regression co-relation ≈ 0.85 | Prediction less accurate; no hybrid optimization; dataset size not specified; no detailed cross-model comparison |
