A practical hybrid NNGA system for predicting the compressive strength of concrete containing natural pozzolan using an evolutionary structure

•NNGA system is faster and performs better than NN model only.•NNGA system can predict the compressive strength at any age with control of concrete mix design.•NNGA system can be used to explore the effect of the concrete mixture and age on the compressive strength.•A new graphical user interface is implemented to facilitate the use of the new NNGA prediction system. Many researchers are interested in predicting the concrete compressive strength, resulting in quite a few linear and nonlinear regression equations. Alternatively, other models have been developed to produce more sophisticated systems by applying soft computing techniques, the majority of which have rarely been used beyond classic problems, such as function optimization or approximation by genetic algorithms (GAs), or neural networks (NNs). Our study proposes an evolutionary structure with a more complex NN in order to achieve the full potential of these techniques, which the genetics of neural systems promises to do. It consists of integrating a GA to optimize the connection weights for each neuron of an NN developed previously. The idea behind this combination is to develop an NNGA model prediction of the compressive strength of concrete containing natural pozzolan. Model learning and testing were first performed based on the back-propagation algorithm. Then, the model was optimized using the proposed evolutionary structure based upon GA. More than 400 experimental data collected from past studies were used in building this model. The hybrid NNGA model was compared with NN model using the same architecture, show that the NNGA is more performant and better than NN alone. The proposed hybrid model was also experimentally validated, very acceptable results with a high correlation coefficient R2 equal to 0.93, yielding comparable results to those obtained by the ACI 209-08 and CEB-FIP models with R2 values equal to 0.95 and 0.96, respectively. However, it can help to predict the compressive strength of a specified concrete mix at any age without knowing in prior the 28days’ compressive strength of this given concrete as it is the case in ACI 208-09 and CEB-FIB Codes. The main feature of this system is its flexibility to reduce significantly the scale of the experiment using a system graphical user interface.