Reliability Assessment of Optimal Extremely Randomized Tree on the Uplift Resistance of Circular Anchors in Anisotropic Clay

<span style="color: rgb(51, 51, 51); font-family: ltr-font; text-align: justify;">An anchor that can withstand an uplift load is one of the necessary options for the offshore raft. The inquiry evaluates the efficacy of the provided models in predicting the uplift resistance factor (F</span><span style="font-size: 9.75px; line-height: 0; position: relative; vertical-align: baseline; bottom: -0.25em; color: rgb(51, 51, 51); font-family: ltr-font; text-align: justify;">c</span><span style="color: rgb(51, 51, 51); font-family: ltr-font; text-align: justify;">=q/(S</span><span style="font-size: 9.75px; line-height: 0; position: relative; vertical-align: baseline; bottom: -0.25em; color: rgb(51, 51, 51); font-family: ltr-font; text-align: justify;">uTC0</span><span style="color: rgb(51, 51, 51); font-family: ltr-font; text-align: justify;">+ρH)) of circular anchors in clays which are both anisotropic and heterogeneous. The statistical findings from prior scientific research have been utilized for training machine learning algorithms. A database included 720 dataset samples. The system’s learning, validating, and testing phases specifically used the dataset's learning (70%), validation (15%), and assessment (15%) sets. Extremely Randomized Tree (ERT) was developed in order to achieve this objective. Since the ERT are connected to the Pelican Optimization Algorithm (POA) and Manta Ray Foraging Algorithm (MRFA) procedures to identify the proper set, hyperparameters are crucial to this experiment. The F</span><span style="font-size: 9.75px; line-height: 0; position: relative; vertical-align: baseline; bottom: -0.25em; color: rgb(51, 51, 51); font-family: ltr-font; text-align: justify;">c</span><span style="color: rgb(51, 51, 51); font-family: ltr-font; text-align: justify;">&nbsp;is estimated using three dimensionless parameters: the normalized outcomes parameter, the embedment ratio, the strength ratio, and the anisotropy ratio. Evidence indicates that both ERT(MRFA) and ERT(POA) can reliably forecast F</span><span style="font-size: 9.75px; line-height: 0; position: relative; vertical-align: baseline; bottom: -0.25em; color: rgb(51, 51, 51); font-family: ltr-font; text-align: justify;">c</span><span style="color: rgb(51, 51, 51); font-family: ltr-font; text-align: justify;">. In comparison to the ERT(POA) approach, the R</span><span style="font-size: 9.75px; line-height: 0; position: relative; vertical-align: baseline; top: -0.5em; color: rgb(51, 51, 51); font-family: ltr-font; text-align: justify;">2</span><span style="color: rgb(51, 51, 51); font-family: ltr-font; text-align: justify;">&nbsp;values for the learning, validation, and evaluation phases of the ERT(MRFA) methodology exhibited accuracy levels of 0.9806, 0.9849, and 0.9808, with corresponding improvement values of 1.24483%, 0.93242%, and 1.32398%. The ERT(POA) attained the highest R</span><span style="font-size: 9.75px; line-height: 0; position: relative; vertical-align: baseline; top: -0.5em; color: rgb(51, 51, 51); font-family: ltr-font; text-align: justify;">2</span><span style="color: rgb(51, 51, 51); font-family: ltr-font; text-align: justify;">&nbsp;values of 0.9928 for training, 0.99401 for validation, and 0.9938 for evaluation. The variance % value, logical analysis, evaluation criteria, and scoring system were applied to analyze the credibility and dependability of each model. In accomplishing its main goal, the ERT(POA) model outperforms the alternative model.</span>