Assessing and forecasting the value of ultra-high-performance concrete using both individual and hybrid schemes with gaussian process regression

The different properties, proportions, and types of the constituents used individually rest at the base of Ultra-High-Performance Concrete (UHPC) performance. Again, machine learning techniques are deemed particularly important to unlock this complex relationship empirically. In this respect, Gaussian Process Regression (GPR) has emerged as one of the most practical Machine Learning (ML) techniques for its superior performance in building a regression model that correlates experimental data. Notably, these models turn out to be impressively accurate and successfully replicate the results of the experiments, further indicating how well GPR works in predicting UHPC behaviors based on input data. The authors combined two meta-heuristic schemes called the Northern Goshawk Optimization (NGO) and Zebra Optimization Algorithm (ZOA) in this study that are used for improving forecast accuracy. This amalgamation gives three hybrid models: GPZO, GPNG, and GPR. The GPZO model stood out predominantly with excellent R² values, having a remarkable score of 0.996 and an ideal RMSE of 2.437 during training. Those indicate the great generalization and estimation capabilities against alternative models developed throughout this study. In other words, sophisticated machine learning through GPR added to the empirical experience, and the intentional incorporation of ZOA and NGO led to improved prediction accuracy regarding the projection of UHPC compressive strength. Overall, the achieved GPZO model displays the best performance among the schemes developed within this investigation.