Abstract
This study describes the prediction and development of a new mathematical model for two parameters of rheology i.e., plastic viscosity (PV) and yield stress (YS) by the application of a novel machine learning algorithm gene expression programming (GEP). An extensive database is established from the experimental results of the previous studies and the six significant input parameters i.e., cement, sand, water, small size coarse gravels, medium size coarse gravels, and superplasticizer were identified most influential parameters affecting the rheological properties of concrete by several trials analyses and chosen as inputs for modeling. The developed GEP mathematical models for both output parameters (PV and YS) showed a strong correlation (R2 of 0.978 and 0.998) with the experimental dataset. Results show that, once the GEP model is precisely trained and its hyperparameters (number of chromosomes, head size, and number of genes) are meticulously optimized, it produces a highly efficient prediction for both rheological parameters. Moreover, the performance index factor with values of 0.0356 and 0.00712 for YS and PV depicts the higher efficiency and predictability of the developed mathematical models. Statistical and external linear and non-linear validation checks demonstrated the high precision, strong predictability, and generalization capacity of developed models for both parameters.