Table 1.

Summary of previous ML models in predicting RCDBs shear strength.

Reference	Category (number)*	Models: Statistical criteria
Wakjira10	WOR (371)	GP: $V_{\mathit{th}}=0.0456f_c^{\prime 0.619}{\rho }_l^{0.411}{\left(\frac{a}{d}\right)}^{-0.874}{b}_{w}d$ μ = 0.82, COV = 0.305
Ashour4	All (141)	GP: $V=b_{w}h\sqrt{f_c^{\prime }}\left[\left(-,4.56,+,1.68,\frac{a}{d}\right),{\rho }_l^2,+,\left(2.45,+,0.1,{\left(\frac{a}{d}\right)}^2,-,1.16,\frac{a}{d},+,3.12,{\rho }_t\right),{\rho }_l,+,0.3,{\rho }_{\mathit{hw}},+,0.4,{\rho }_{\mathit{vw}}\right]$ μ = 1.11, Std = 0.21
Shahnewaz9	All (381)	GA: $V_u=b_{w}h{f}_c^{\prime }\left[\frac{2}{5},-,\frac{1}{4},{\left(\frac{a}{d}\right)}^{0.23},+,0.85,{\left({\rho }_l,{\rho }_{\mathit{hw}},{\rho }_{\mathit{vw}}\right)}^{0.1},-,\frac{3}{5},{\left(\left(\frac{a}{d}\right),{\rho }_{\mathit{hw}},{\rho }_{\mathit{vw}}\right)}^{\frac{1}{16}},-,200,{\left(\left(\frac{a}{d}\right),{\rho }_l,{\rho }_{\mathit{hw}},{\rho }_{\mathit{vw}}\right)}^{2.65}\right]$ μ = 0.99, CoV = 0.232 GA: $V_u=b_{w}h{f}_c^{\prime }\left[1.74,-,2,{\left(\frac{a}{d}\right)}^{0.044},+,0.5,{\rho }^{0.14}\right]$ μ = 1.01, CoV = 0.257
Cheng5	All (106)	EMARS, BPNN, RBFNN, SVM. EMARS is the best model. Grid search with cross-validation EMARS: training MAPE = 5.67, R2 = 0.989, testing MAPE = 5.887, R2 = 0.973
Feng6	All (271)	DT, SVM, ANN, RF, AdaBoost, GBRT, XGBoost. XGboost is the best model. Grid search with cross-validation XGboost: Training R2 = 0.999, MAPE = 0.74, testing R2 = 0.928, MAPE = 10.44% (overfitting)
Hameed11	All (271)	LWR, RF, MLR, ELM. LWR is the best model. Grid search with cross-validation LWR: Training: RMSE = 22.563, MAE = 13.249, a20-index = 98.89, testing: RMSE = 57.776, MAE = 33.933, a20-index = 85.87
Liu8	All (267)	LR, SVR, ANN, RF, XGBoost, NGBoost using Bayesian optimisation technique. NGBoost is the best model NGboost: R2 = 0.9045, RMSE = 38.7976 kN
Tiwari7	All (271)	DT, SVR, RF, GB, Adaptive boosting, XGBoost, voting regression. XGboost is the best model. Grid search with cross-validation XGboost: Training R2 = 0.999, MAPE = 0.78, RMSE = 1.45 kN, testing R2 = 0.928, MAPE = 9.79, RMSE = 47.76 (overfitting). μ = 1.00, CoV = 6.38%
Nguyen3	All (518)	LR, ANN, SVR, DT, GPR, XGBoost using Bayesian optimisation technique. GPR is the best model GPR: Training R2 = 0.99, MAE = 12.77, RMSE = 18.84 kN, validation R2 = 0.89, MAE = 41.72, RMSE = 71.06 kN, testing R2 = 0.94, MAE = 38.44, RMSE = 63.38
Ma2	All (457)	kNN, DTM RFM GBDT, CatBoost, XGboost. XGboost is the best model. Grid search with cross-validation XGboost: Training R2 = 0.992, MAE = 0.148, RMSE = 0.26, testing R2 = 0.917, MAE = 0.531, RMSE = 0.777 WOR*: μ = 1.03, Std = 0.128, WVR*: μ = 1.005, Std = 0.073, WHR*: μ = 1.003, Std = 0.077, WVHR*: μ = 1.01, Std = 0.084
This study	All (840)	CATBoost: Training: μ = 1.005, CoV = 0.062, a20-index = 0.9894, MAPE = 4.41, RMSE = 36.8 kN Testing: μ = 1.026, CoV = 0.141, a20-index = 0.899, MAPE = 9.32, RMSE = 160.9 kN SR: (WOR)* μ = 1.003, CoV = 0.207, a20-index = 0.68, MAPE = 16.80, RMSE = 115.9 kN SR: (WWR)* μ = 1.004, CoV = 0.192, a20-index = 0.78, MAPE = 13.70, RMSE = 196.7 kN

*WOR and WWR stand for without web reinforcement and with web reinforcement cases.