Compressive strength of Foamed Cellular Lightweight Concrete simulation: New development of hybrid artificial intelligence model

Accurate prediction of compressive strength (f_c) is one of the crucial problems in the concrete industry. In this study, novel self-adaptive and formula-based model called Multivariate Adaptive Regression Splines optimized using Water Cycle Algorithm (MARS-WCA) is proposed for modeling f_c based on mixture proportion. The proposed predictive model is validated against several benchmark models including Multiple Linear Regression (MLR), Artificial Neural Network (ANN), Support Vector Regression (SVR) and standard MARS model. 418 experimental datasets are collected from the open-source literatures to calibrate and validate the computational intelligence models. The best subset regression procedure is conducted based on different forms of combinations using Mallow's coefficient to specify the effective variables influencing the f_c of Foamed Cellular Lightweight Concrete (FCLC). The applied MARS-WCA model is evaluated with the external validation and uncertainty analysis. It is found that foam, sand, binder, water to cement ratio, sand to cement ratio and age of specimens are the most essential predictors to provide the minimum Mallow's coefficient value. In quantitative terms, MARS-WCA attained (NSE = 0.938) and that reporting an enhancement of FCLC compressive strength prediction capability over the MLR, ANN, radial basis function-SVR, polynomial-SVR and MARS by 39.4%, 9.2%, 9.6%, 41.7% and 4.7% in term of Nash-Sutcliffe efficiency indicator. Overall, the proposed self-adaptive MARS-WCA model demonstrated a robust and significant data-intelligence mode for FCLC compressive strength prediction compared with the benchmark models and experimental formulations.