Abstract:
Compressive strength of high early strength is related non-linear with its components. In order to effectively utilize the high early strength concrete, compressive strength estimation is necessary. Nowadays, compressive strength prediction is done by a lot of researchers using machine learning tools with acceptable accuracy and performance. Therefore, in this research, an integrated framework of genetic algorithms and artificial neural networks was proposed. The compiled data consisting of eleven input and one output variable was used to train the model. The developed model has shown a high accuracy with a correlation coefficient of around 0.98 and a mean absolute error close to 3. Feature importance demonstrated that cement and water-binder ratio are the top two candidates contributing to the model whereas partial dependence plot analysis showed output variation with inputs. The comparison with other in-practiced machine learning techniques showed that the developed model has the highest performance. In addition, the lab-scale experiments further provided an evaluation of the model and displayed those results from the model and actual are close to the 5 percent error.
