Optimizing compressive strength prediction of sustainable concrete using ternary-blended agro-waste ash, sugarcane bagasse ash, and rice husk ash with soft computing techniques

This study examines the properties of sustainable concrete produced by partially substituting ordinary Portland cement (OPC) with agricultural waste, including sugarcane bagasse ash (SBA) and rice husk ash (RHA). This research investigated the effects of these substitutions both individually and in combination while completely replacing river sand with artificial sand. A total of 18 M-30 grade concrete mix proportions were prepared and tested to evaluate their compressive strength, utilizing 164 experimental datasets. To predict the compressive strength of sustainable concrete, advanced machine learning techniques have been employed. Hybrid models were developed by optimizing the hyperparameters of artificial neural networks (ANNs) using four metaheuristic algorithms: the Harris hawk optimizer (HHO), particle swarm optimization (PSO), gray wolf optimizer (GWO), and the slime mold algorithm (SMA). The models' prediction performance was assessed using eight metrics, including R-2, VAF, MAE, and RMSE, alongside comparative analyses based on a comprehensive measure (COM), performance strength criteria (PSC), scatter plots, and residual error histograms. The results demonstrate that the ANN-HHO model consistently outperforms other hybrid models. In the training phase, ANN-HHO achieved the highest performance metrics (R-2 = 0.9606, VAF = 96.0363), with similar excellence in the testing phase (R-2 = 0.9580, VAF = 95.8005). Additionally, ANN-HHO exhibited the lowest error metrics (MAE = 0.0339, RMSE = 0.0493 in training; MAE = 0.0384, RMSE = 0.0509 in testing). It also achieves the best COM (0.712) and PSC (0.940) values, confirming its superior accuracy and robustness compared with the ANN-GWO, ANN-PSO, and ANN-SMA models. The error residual histogram further supported the model's unbiased predictions and smaller standard deviation. Experimental validation verified the efficacy of the ANN-HHO model as a reliable alternative tool for accurately predicting the compressive strength of sustainable concrete incorporating agricultural waste. These findings highlight the potential of machine learning-assisted approaches in advancing sustainable construction practices.