Assessment of ultimate bearing capacity of rock-socketed piles using hybrid approaches

In geotechnical engineering, accurately estimating the ultimate bearing capacity ( \({Q}_{u}\) $Q_u$ ) of rock-socketed piles remains a crucial challenge. This study introduces a methodology that integrates advanced optimization algorithms, specifically Dandelion Optimization (DO) and the Crystal Structure Algorithm (CryStal), with Support Vector Regression (SVR) to enhance predictive capabilities. Three distinct models—SVDO, SVCS, and a hybrid SVR model—are developed through this integration. The core of this predictive framework is SVR, known for its efficacy in capturing intricate non-linear relationships between input variables and the ultimate bearing capacity of rock-socketed piles. To improve predictive accuracy, DO strategically adjusts hyperparameters to emulate the growth and dispersal patterns of dandelion seeds, while CryStal delicately optimizes SVR parameters inspired by crystalline atomic structures. The resulting models offer valuable insights for precisely predicting the ultimate bearing capacity of rock-socketed piles in geotechnical engineering. Among these, SVCS stands out with an exceptional R² value of 0.997, indicating an outstanding fit to the data, and the lowest Root Mean Squared Error (RMSE) at 930.7, underscoring its unparalleled predictive accuracy. In conclusion, this study presents an innovative approach within geotechnical engineering for the precise estimation of the ultimate bearing capacity of rock-socketed piles. The insights gained contribute significantly to considerations of stability and safety in construction projects, emphasizing a multidisciplinary approach beyond artificial intelligence.