Developing a Novel Machine Learning Method to Predict the Compressive Strength of Fly Ash Concrete in Different Ages

Abstract

Estimating the compressive strength of concrete before fabricating, has been one of the most important challenges because designing a mixture proportion by experimental methods needs expert workers, consumes energy, and wastes materials. Therefore, in this study, the influences of materials and the age of samples on the compressive strength of fly ash concrete are investigated, and a novel method for predicting the compressive strength is presented. To this end, the water cycle algorithm and genetic algorithm are utilized, and their outcomes are compared with the classical regression models. Various performance indicators are used to gauge the accuracy of the models. By analyzing the results, it is comprehended that the water cycle algorithm is the most accurate model according to all performance indicators. Besides, the outcomes of the water cycle algorithm and genetic algorithm are by far better than those of classical methods. The mean absolute error of water cycle algorithm, genetic algorithm, linear regression, partial-fractional regression, and fractional regression are 3.01, 3.12, 5.47, 9.70, and 5.37 MPa for training data and 2.90, 3.44, 5.47, 9.70, and 5.37 MPa for testing data respectively. Furthermore, the water cycle algorithm is the only algorithm whose mean absolute error of testing data is less than that of training data. At last, it was concluded that the mixture with less than 35% fly ash (weight of the binder) had maximum amounts of compressive strength. Also, the compressive strength of concrete decreased significantly as the amount of fly ash increased more than this definite level.