Modelling of compressive strength of geopolymer paste, mortar and concrete by optimized support vector machine

Abstract

To date, many artificial intelligence-based techniques have been conducted to predict compressive strength of concrete specimens. However, modified models by metaheuristic optimization algorithms to present approaches with higher performance may be of special interest. In this work, support vector machine (SVM) is utilized to predict compressive strength of geopolymers, cement-free eco-friendly construction materials. Parameters of SVM are sometimes hard to be found especially in the case of complex models. Therefore, task of finding these parameters was followed by five different well-known optimization algorithms including genetic algorithm (GA), particle swarm optimization algorithm (PSOA), ant colony optimization algorithm (ACOA), artificial bee colony optimization algorithm (ABCOA) and imperialist competitive algorithm (ICOA). Results of these five hybrid models were compared by a model using just SVM, and other traditional artificial intelligence techniques namely artificial neural networks (ANNs) and adaptive neuro-fuzzy interfacial systems (ANFIS). A total number of 1347 data were collected from the literature for modelling. It was suggested that hybrid models can be appropriately used for modelling of compressive strength of geopolymer paste, mortar and concrete specimens. By evaluating the proposed models through their coefficient of determination and errors, it was concluded that ICOA and GA are more suitable to optimize parameters of SVM for predicting compressive strength of the considered geopolymers. Additionally, ANN model was remained as one of the simplest approaches which can be used with reasonable accuracy for the problem of this paper.