Enhancing soil pile-bearing capacity prediction in geotechnical engineering using optimized decision tree fusion

Geotechnical engineering places a high priority on accurately predicting the soil pile-bearing capacity (Pu). Machine learning (ML) must be crucial to achieving this precision. A ground-breaking method combines the adaptable Decision Tree (DT) technique with cutting-edge optimization algorithms like Tasmanian Devil Optimization (TDO) and Gold Rush Optimizer (GRO). This fusion produces three distinct models: DTTD, DTGR, and a hybrid DT model. DT, which is renowned for its ability to capture complex non-linear relationships between input variables and soil P_u, is at the center of this predictive framework. The effectiveness of DT, however, depends on accurate hyperparameter tuning and the choice of a suitable optimization approach, which is precisely where TDO and GRO come into play. GRO strategically modifies hyperparameters to simulate the development and dispersal of dandelion seeds, improving the DT’s ability to predict the future. On the other hand, TDO uses the crystalline atomic structure as inspiration to optimize DT parameters precisely. The power of DT and these state-of-the-art optimization methods are combined in this synergy to produce three distinctive models. DTTD stands out among these models as the top performer, with an exceptional R² value of 0.996, indicating an excellent fit for the data. Additionally, it has the lowest RMSE, which stands at 23.221, and highlights its unmatched predictive accuracy. Thus, the DTTD model represents a novel and highly successful strategy for precisely forecasting soil P_u.