A comparison of novel hybrid ensemble learners to predict the compressive strength of green engineering materials: a case of concrete composed of rice husk ash

The use of ensemble learning (EL) has grown due to its ability to enhance precision in predictions compared to typical machine learning (ML) algorithms. EL-based approaches are expected to be more accurate and efficient in simulating green engineering materials like sustainable concretes. In this study, three novel EL methods predict the compressive strength (CS) of concrete containing rice husk ash (CCRHA) using hybridizations of adaptive boosting (AdaBoost), gradient boosting regression trees (GBRT), random forests (RF), and teaching-learning-based optimisation algorithms (TLBO). Data for the CS of CCRHA (1212 instances) were collected from 42 published journal papers. Input variables included the amount of water, cement, fine aggregate, coarse aggregate, rice husk ash, superplasticizer, and curing time. Generally, GBRT-TLBO outperformed other approaches in testing with R2, RMSE, and MAE values of 0.976, 3.299, and 2.419 MPa, respectively. GBRT-TLBO models had superior capabilities in the testing phase with interquartile ranges of absolute error boxes equal to 1.0 MPa. A comparison with literature models showed that GBRT-TLBO was correctly tuned. Sensitivity analysis indicated cement and age as the most significant variables, while RHA was the least significant. Surface plots from parametric studies identified the interaction of input variables with concrete CS.