Supervised and unsupervised machine learning techniques for predicting mechanical properties of coconut fiber reinforced concrete

In the quest for enhancing the crack resistance of concrete structures, various technologies, and additives have been successfully employed, with fiber reinforcing emerging as an effective technique for various construction activities. This study explores the predictive capabilities of soft computing approaches, including linear regression (RF), support vector machine (SVM), bagging with SVM, additive regression with SVM, M5P, artificial neural network (ANN), and random forest (RF), in forecasting mechanical properties such as compressive strength (CS), split tensile strength (STS), and flexural strength (FS) of coconut fiber-reinforced concrete (CFRC). The dataset, amalgamated from experimental findings and pertinent research papers, serves as the foundation for model evaluation using five statistical indices: coefficient of correlation (CC), root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and scatter index (SI). Notably, the SVM model excels, exhibiting a CC of 0.9523, RMSE of 3.3362, MAE of 2.0071, MAPE of 0.1306, and SI of 0.1309 for predicting CS. Conversely, RF yields superior results in predicting STS, boasting a CC of 0.9841, RMSE of 0.2748, MAE of 0.2243, MAPE of 0.1411, and SI of 0.1187. Moreover, RF outperforms in predicting FS for CFRC. The comprehensive analysis indicates the reliability of all applied techniques in predicting the CS, STS, and FS of concrete. Additionally, a sensitivity analysis, based on the best-performing model, underscores the significant impact of curing time on the predictive accuracy of the models. The study emphasizes the critical role of controlling factors like CF content and curing time in coconut-eforced concrete mixtures to enhance mechanical properties. © The Author(s), under exclusive licence to Springer Nature Switzerland AG 2024.