Exploring the potential of soft computing for predicting compressive strength and slump flow diameter in fly ash-modified self-compacting concrete

Self-compacted concrete (SCC) is one of the special types of concrete. The SCC represents one of the most significant developments in concrete technology over the previous two decades. It can compact itself using its weight without requiring vibration due to its excellent fresh characteristics, which allow it to flow into a uniform level under the impact of gravity. Since cement manufacturing is one of the largest contributors to CO 2 gas emissions into the atmosphere, fly ash (FA) is used in concrete as a cement replacement. Currently, FA-modified SCC is widely utilized in construction. This research aimed to study the potential of soft computing models in predicting the compressive strength (CS) and slump flow diameter (SL) of self-compacted concrete modified with different fly ash content. Hence, two databases were created, and relevant experimental data was collected from previous studies. The first database consists of 303 data points and is used to predict the CS. The second database predicts the SL and contains 86 data points. The dependent parameters are the CS, which varies from 9.7 to 79.2 MPa, and the SL, which varies from 615 to 800 mm. The identical five independent parameters are available in each database. The ranges for CS prediction are water-to-binder ratio (0.27–0.9), cement (134.7–540 kg/m 3 ), sand (478–1180 kg/m 3 ), fly ash (0–525 kg/m 3 ), coarse aggregate (578–1125 kg/m 3 ), and superplasticizer (0–1.4%). The data ranges for the SL prediction, on the other hand, are as follows: water-to-binder ratio (0.26–0.58), cement (83–733 kg/m 3 ), sand (624–1038 kg/m 3 ), fly ash (0–468 kg/m 3 ), coarse aggregate (590–966 kg/m 3 ), and superplasticizer (0.1–21.84%). Each database has developed three models for the prediction: full-quadratic (FQ), interaction (IN), and M5P-tree models. Each database is divided into two groups, with training comprising two-thirds of the total data points and testing containing one-third. As a result, 202 training data and 101 testing data are in the first database. The other database consists of 57 data points for training and 29 for testing. Various statistical tools are used to evaluate the performance of each proposed model, such as R 2 (correlation of coefficient), RMSE (root mean squared error), SI (scatter index), MAE (mean absolute error), StDev, OBJ (objective value), a-20 index, and Z-score. The results showed that the FQ and IN models have the highest accuracy and reliability in predicting the compres