Comparison of Machine Learning Models for Predicting Environmental Risk Associated with Coastal Waste at a Global Level (#981)

Authors

Fernandez Vasquez, Richard Fernando

Abstract

Machine learning models are key tools in coastal environmental risk management, as they allow for the identification of critical areas, optimize waste management, and support the development of more effective and sustainable conservation policies globally. This research aimed to compare machine learning models for predicting the environmental risk associated with coastal waste globally, with the goal of identifying the most suitable model and guiding the formulation of coastal conservation policies. The research used a database of 165 countries with varying levels of environmental risk associated with coastal waste. The data were divided into a training sample (80%) and a validation sample (20%). The performance of five Machine Learning models —Random Forest, Gradient Boosting, XGBoost, LightGBM and CatBoost— was evaluated in predicting the probability of environmental risk associated with coastal waste at a global level, with the Random Forest model showing the best performance, with an accuracy of 0.5455, recall of 0.8000, F1-score of 0.6486, area under the ROC curve of 0.6852 and Gini index of 0.3704, demonstrating the greatest capacity for discrimination and predictive accuracy.