Predicting the uniaxial compressive strength of oil palm shell lightweight aggregate concrete using artificial intelligence‐based algorithms

• Published in: Structural concrete : journal of the FIB Vol. 23; no. 6; pp. 3631 - 3650
• Main Authors: Zhu, Weixiang, Huang, Lihua, Mao, Liudan, Esmaeili‐Falak, Mahzad
• Format: Journal Article
• Language: English
• Published: Weinheim WILEY‐VCH Verlag GmbH & Co. KGaA 01.12.2022; Wiley Subscription Services, Inc
• Subjects: Algorithms; Artificial intelligence; Compressive strength; Concrete aggregates; Concrete industry; extreme gradient boosting; Lightweight; lightweight aggregate concrete; multivariate adaptive regression spline; oil palm shell; Optimization; optimization algorithms; Palm oil; Particle swarm optimization; Performance prediction; Regression models; Swarm intelligence; Trigonometric functions; uniaxial compressive strength
• ISSN: 1464-4177; 1751-7648
• DOI: 10.1002/suco.202100656

Because natural coarse aggregates were depleting rapidly, concrete industry has been trended toward substitute aggregates from industrial by‐products or waste. One of the waste materials is oil palm ash (OPS), which is widely generated in the processing of palm oil in the tropics. Concretes made with OPS to estimate the compressive strength (CS) is cost and time consuming. This study aims to propose novel hybrid models by concepts of extreme gradient boosting (XGB) model optimized with different optimization algorithms such as sine–cosine algorithm, multiverse optimization algorithm (MVO), and particle swarm optimization for predicting the uniaxial CS (UCS) of oil palm shell lightweight aggregate concrete (OPS). Also, the multivariate adaptive regression spline model is also developed to present a meaningful relationship between input and output variables. To this aim, a data set containing data samples for concrete made with OPS was gathered from the published literature. Results show that all models have acceptable performance in predicting the UCS, representing the admissible correlation between observed and predicted values and models' robustness. In the training step, the value of R2, the root mean square error, and the variance accounted factor for MVO–XGB are 0.9713, 1.5777, and 97.129. These values in testing phase are 0.9019, 2.6786, and 89.158. Therefore, the MVO–XGB model outperforms others, and the results demonstrate the ability of the MVO algorithm to determine the optimal value of XGB parameters.