Predicting compressive strength of ultra-high-performance concrete using Naive Bayes regression in novel approaches

The unique properties, quantities, and varieties of Ultra-High-Performance Concrete's (UHPC) constituent materials are intimately related to the material's compressive strength. Machine Learning (ML) algorithms are essential for understanding this complex relationship empirically. The Naive Bayes Regression (NBR) algorithm stands out among these as being exceptionally skilled at creating a predictive model that matches experimental datasets. The fact that these models are remarkably accurate and closely match experimental results is particularly noteworthy. This demonstrates the ability of NB to forecast UHPC behavior based on input parameters. This study adopts a novel strategy by combining 2 meta-heuristic algorithms, the Weevil Damage Optimization Algorithm (WDOA), and the Gold Rush Optimizer (GRO), as well as the ensemble of these 2 optimizers, to increase the accuracy of these predictions. 4 models were created as a result of this integration: NBGR (combination of NB and GRO), NBWD (combination of NB and WDOA), NBWG (combination of NB and ensembled WDOA-GRO) and NB (single model). Among them, the NBWD model stands out due to its exceptional R² values of 0.987, along with an ideal RMSE during the testing phase of 5.361. Compared to previous models developed in this study, these indicators demonstrate the model's unparalleled generalizability and predictive potential. Essentially, the combination of advanced ML techniques through NB, WDOA, and GRO, along with thoughtful integration, improves predictive brilliance in the projection domain.