Data-driven optimization with PSO improved Random Forest regression and sensitivity analysis for the strength of blended concrete

The physical properties of concrete depend on the type of supplementary cementitious materials (SCMs); thus, its strength must be evaluated for specific purposes. In this study, the Random Forest regression (RFR) machine learning (ML) algorithm was used to estimate the compressive strength (CS) of blended concrete (BC) with fly ash (F_Ash). The hyperparameters of the RFR model were optimized using Particle Swarm Optimization (PSO) to enhance predictive performance. The optimized RFR model served as a surrogate model, and PSO found optimal input parameters for the best response. Key inputs included cement, GGBS, fine aggregate (FA), coarse aggregates (CA), fly ash (F_Ash), water content, superplasticizer and curing days, with CS as the output. Performance evaluation used indices, such as MAE, MAPE, MSE, RMSE, MBE, R2, a20-index to assess accuracy. Sensitivity analysis showed the relationship between inputs and CS, highlighting the impact of F_Ash and other parameters on CS prediction. The model achieved an RMSE of 2.005 and an R2 of 0.9858 for CS. The optimized response was confirmed at 71.42 MPa, with optimized input parameters: Cement = 365.94 kg/m3, GGBS = 263.184 kg/m3, F_Ash = 148.610 kg/m3, W = 123.907 litres, SP = 23.58 kg/m3, CA = 820.788 kg/m3, FA = 734.118 kg/m3, and days = 166.710.