Incorporating knowledge graph and multi-model stacking ensemble learning for prediction of fines for illegal fishing

Abstract

Illegal fishing activities have a significant threat to global marine resource management, inflicting severe damage on marine ecosystems, disrupting legal fisheries economies, and hindering biodiversity conservation efforts. Research on Illegal, Unreported, and Unregulated (IUU) fishing activities have gained substantial attention globally, focusing on identification, monitoring, prevention, and policy formulation. However, limited efforts have been directed toward the prediction of fines and the assessment of legal repercussions. This gap hampers the timely accurate evaluation of the economic impacts of IUU fishing behaviors and undermines the full deterrent potential of legal penalties. Therefore, this paper proposes an integrated method combining knowledge graph and multi-model stacked generalization, aiming to enhance the accuracy of fines prediction for illegal fishing activities. Experimental results demonstrate that the proposed model significantly enhances prediction accuracy, interpretability, and stability compared with basic machine learning models, including eXtreme Gradient Boosting (XGBoost), Gradient Boosting Machines (GBM), K-Nearest Neighbor(K-NN), Categorical Boosting (CatBoost), and Random Forest(RF). This study provides a new technical guidance for the prediction of fines for illegal fishing activities, contributing significantly to enhancing the efficiency and effectiveness of fishery law enforcement.