Machine learning-based prediction of unconfined compressive strength and contaminant leachability in dredged contaminated sediments for land reclamation projects.

This research investigates the application of machine learning techniques for predicting unconfined compressive strength (UCS) and contaminant leachability in dredged contaminated sediments (DCS) with implications for land reclamation projects. Traditionally, determining these parameters has been challenging, costly, and time-consuming, hindering efficient project planning and execution. Therefore, this study evaluated the efficacy of two machine learning models, namely extreme gradient boosting (XGBoost) and decision tree (DT), in improving prediction accuracy and reducing the need for resource-intensive testing procedures. The models were constructed using 165 data samples and 6 input parameters. The models' generalizability and predictive performance were evaluated using Monte Carlo and K-fold cross-validation approaches. Results indicate that the XGBoost model outperforms DT, exhibiting superior prediction accuracy and consistency with best typical metrics such as higher adjusted R-squared (Adj. R) and lower root mean square error (RMSE) and mean absolute error (MAE) values. The sensitivity analysis of models shows that the ground granulated blast furnace slag (GGBS) content has a significant impact on the prediction of UCS, whereas the zinc concentration level (Z) has a significant effect on the leachability of zinc. These findings demonstrate the ability of machine learning to refine prediction algorithms for DCS performance, allowing for more efficient and cost-effective land reclamation initiatives.