Unveiling the potential of an evolutionary approach for accurate compressive strength prediction of engineered cementitious composites

The different human activities in numerous fields of civil engineering have become possible due to recent development in soft computing. As many researchers have widely extended the use of evolutionary numerical methods to predict the mechanical properties of construction materials, it has become necessary to investigate the performance, accuracy, and robustness of these approaches. Gene Expression Programming (GEP) is a method that stands out among these methods as it can generate highly accurate formulas. In this study, two models of GEP are used to anticipate the compressive strength of engineered cementitious composite (ECC) containing fly ash (FA) and polyvinyl alcohol (PVA) fiber at 28 days. The experimental results for 76 specimens, which are made with ten different mixture properties, are taken from the literature to build the models. Considering the experimental results, four different input variables in the GEP approach are used to arrange the models in two modes: sorted data distribution (SDD) and random data distribution (RDD). Prognosticating the compressive strength values based on the mechanical properties of ECC containing FA and PVA will be possible for the models of the GEP method by using these input variables. The comparison between the experimental results and the results of training, testing, and validation sets of two models (GEP-I and GEP-II), each of which has two distinct distribution modes, is done. It is observed that both modes of RDD and SDD lead to responses with the same accuracy (R-square more than 0.9). Nevertheless, the GEP-I (SDD) model was chosen as the best model in this study based on its performance with the validation data set.