Iceberg-seabed interaction analysis in sand by a random forest algorithm

• Published in: Polar Science
• Format: Article in Journal/Newspaper
• Language: English
• Published: 2022
• Subjects: Iceberg-seabed interaction; Sandy seabed; Random forest; Gradient boosting model; Support vector regression; Arctic; Iceberg*; Polar Science
• Online Access: https://nipr.repo.nii.ac.jp/?action=repository_uri&item_id=16985; http://id.nii.ac.jp/1291/00016854/

Iceberg-seabed interaction that threatens subsea pipelines and structures is a challenging and costly engineering design aspect of Arctic offshore infrastructures. In this study, the sub-gouge soil deformation in the sand along with the keel reaction forces was simulated using Random Forest (RF) as a strong machine learning (ML) model and compared with the Gradient Boosting Model (GBM), and Support Vector Regression (SVR) as other alternatives. Nine RF models were built based on the most influential parameters and the best model was identified by performing a sensitivity analysis. The study showed that the proposed RF model outperformed its counterparts and proved to be a cost-effective and reliable alternative to assess the iceberg-seabed interaction in the sand, particularly at the early stages of the projects, where a fast and accurate estimation is required for planning the construction methodologies, logistics, and the scope of detailed engineering.