Estimating unconfined compressive strength using a hybrid model of machine learning and metaheuristic algorithms

Unconfined Compressive Strength (UCS) is one of the most important mechanical properties in geomechanics and is crucial for reliable geo-mechanical modeling of geological formations. Traditional methods for determining UCS involve a great deal of laboratory testing on core samples, which can be very expensive, or collecting a large amount of data from well logs, which could be equally burdensome. Recently, with increasing demands on real-time, efficient UCS predictions in industries dealing with construction, mining, civil engineering, and even petroleum exploration, the need to realize innovative ways has grown. This paper, therefore, embarks on the introduction of a state-of-the-art machine learning framework that embeds the Multi-Layer Perceptron architecture with advanced meta-heuristic optimization techniques, namely Beluga Whale Optimization and Black Widow Optimization, to predict unconfined compressive strength with high accuracy. That is, unlike ANNs, which on occasions easily get bogged down in difficulties of parameter optimizations and slow convergence, it can enable fast and sure predictions, hence its eminently suitable use in real-time decision-making. Employing an extensive dataset compiled from previous scholarly studies, outstanding predictive performance was realized for this model. Therefore, the best-optimized model of MLBO2 was the superior predictor that gave a maximum \({R}^{2}\) $R^2$ of \(0.998\) $0.998$ and minimum RMSE of \(1.309\) $1.309$ in the test phase. These results confirm that the model is efficient in providing highly accurate UCS predictions with immense added advantages over the existing techniques and can contribute much to the geomechanical domains.