Discussion of the article 'Prediction of creep of recycled aggregate concrete using back-propagation neural network and support vector machine'

Abstract

In a recent study, Rong et al.1 investigate the prediction of recycled aggregate concrete (RAC) creep using back-propagation neural network and support vector machine. For this purpose, the authors compiled a database of experimental results on the creep of RAC on which they first tested five analytical RAC creep prediction models2-6 and concluded that the performance of all five models is inadequate, thereby justifying the use of a back-propagation neural network and a support vector machine. The main argument for declaring the performance of the five analytical models inadequate is the analysis of “performance indices” of the correlation coefficient (R), mean absolute error (MAE), mean square error (MSE), and integral absolute error (IAE). The found ranges of values were 0.45–0.55 for R, 0.41–0.64 for MAE, 0.33–0.70 for MSE, and 0.33–0.53 for IAE. Nonetheless, there are errors and uncertainties regarding the study that are pointed out herein, some methodological and some formal.

Postprint (published version)