Estimating compressive strength of modern concrete mixtures using computational intelligence: A systematic review

• Review demystifies use of machine learning in predicting properties of concrete. • Hyperparameters of ML along with their accuracy are critically analyzed and discussed. • Main findings of NL predictions of compressive strength of various concrete types are presented. • Recommendations for best practice are made and needed future research is identified. The mixture proportioning of conventional concrete is commonly established using regression analysis of experimental data. However, such traditional empirical procedures have proven less accurate for modern complex cementitious composites. The lack of robust predictive tools for estimating the mixture composition and engineering properties of novel concretes led to deploying machine learning techniques. Although these versatile computational algorithms have proven successful in diverse applications, their performance is highly dependent on the data structure and appropriate selection of hyperparameters. Therefore, this paper demystifies the use of ML in concrete technology by systematically surveying and critically reviewing ML algorithms employed to predict the compressive strength of modern concrete mixtures. The hyperparameters of various machine learning models along with the achieved accuracy are critically analyzed and discussed. The main findings regarding machine learning predictions of compressive strength for various concrete types are presented, recommendations for best practice are made, and needed future research is identified.