Predicting compressive strength of geopolymer concrete using machine learning

Abstract

The anaconda software required python code in order to run the utilized individual K-nearest neighbor (KNN), random forest regression (RFR), and linear regression (LR) models. The results show that RFR machine learning (ML) technique out of the other utilized models shows the best performance for a used dataset. The findings of this article indicate that the dataset utilized proposed model provides an acceptable algorithm for FACC design and optimization. In the current study of preparation of geopolymer concrete (GPC), relevant variables such as curing, fly ash, calcined clay, added water, super plasticizer, coarse aggregate, quarry stone dust, caustic soda, and water glass were used as input parameters. The ranges, mode, median, standard deviation, and other identifying details were checked using descriptive statistical analysis for the input parameters. The strength due to the compression of FACC GPC was predicted using RFR, LR, and KNN ML techniques, all based on Python coding. The ensemble ML technique, RFR outperformed the individual ML technique, KNN, in terms of prediction. The RFR indicates that the maximum amount of [Formula: see text] is 0.92, and LR provides 0.58, although the KNN was less accurate, with a coefficient of determination of 0.56. The RFR technique’s lower values of errors, mean absolute error (MAE), MSE, and root mean square error (RMSE) yield 1.99, 7.17, and 2.67[Formula: see text]MPa, respectively. The excellent accuracy of the RFR methodology is confirmed by a statistical analysis of errors. Curing temperature, curing hours, molarity of NaOH, and FACC ratio significantly affect the compressive strength (CS) of FACC GPC. The findings indicate that the proposed model provides an acceptable algorithm for FACC design and optimization using RFR among the three combinations of ML methods for a given dataset.