Abstract
Construction labour productivity is affected by a number of factors. In this study, multilayer perceptron neural network (MLPNN), support vector machine (SVM), general regression neural network (GRNN), and multiple additive regression trees (MART) methods were developed to estimate the labour productivity rates of concrete construction activities. Various soft computing techniques were used to examine their respective results to identify the best method for estimating expected productivity. The predictive behaviours of the different techniques are compared with those found in previous productivity studies. The results show that for predicting labour productivity for steel fixing and concrete pouring and finishing, the GRNN model outperforms the other techniques. The GRNN model provided improvements in the root mean square error (RMSE) of 199.41%, 23.21%, and 53.46% over MLPNN, SVM, and MART, respectively, for labour productivity of steel fixing, and 3,311.78%, 681.81%, and 776.68%, respectively, for labour productivity of concrete pouring and finishing. For predicting labour productivity for formwork assembly, the MART method was found to be the superior one, providing improvements in the RAISE of 232.93%, 90.89%, and 28.88% over MLPNN, GRNN, and SVM, respectively.