Predicting the compressive strength of foam concrete: an in-depth investigation employing material analysis and beetle antennae search-random forest modelling

Foam concrete (FC), recognized as a lightweight construction material, has gained attention for its potential to reduce structural dead loads. However, achieving constant strength due to the lack of standardized mix design remains a challenging factor in FC. To streamline this process, researchers proposed indirect estimation methods, and Machine Learning (ML) techniques for predicting the Compressive Strength (CS) of FC. However, their application is difficult for non-linear datasets. Therefore, the study addresses the gap by employing material analysis techniques such as, visualization techniques, such as SHAP (SHapley Additive exPlanations) analysis, feature importance analysis, data exploration analysis, and Partial Dependency Plots (PDP) for model interpretation. From the material analysis it is observed that, addition of 10% silica fume in cement increases the strength by 56% in FC. SHAP analysis identifies cement and Silica Fume (SF) as pivotal contributors. while feature importance analysis and PDP identifies the cement, SF, and water are major contributors of FC. Data exploration highlights the SF as strength enhancer for FC. Later, The Beetle Antennae Search-based random forest regression (BAS-RF) is employed for predicting the CS of FC. The BAS-RF algorithm is used for model training, evaluation, comparison, residual analysis, and ensemble techniques. The BAS-RF model, trained on a diverse dataset, exhibits robustness across varying levels of cement and SF. From BAS-RF Model evaluation a strong predictive performance with R² of 0.9 and RMSE of 1.30 is observed. In a comparative analysis, the BAS-RF model with R² of 0.9 outperforms Support Vector Regression (SVR) with R² of -0.18.