Machine Learning Based Prediction of Coastal Waste Risks Using Global Plastic Waste Production Data

Abstract

Plastic pollution poses a growing threat to coastal ecosystems, marine biodiversity and human activities. Identifying coastal areas at high risk of plastic waste accumulation is important for developing effective waste management strategies. However, assessing this risk requires consideration of multiple factors, including plastic waste generation, waste mismanagement, population pressure and environmental conditions. This study presents a machine learning based approach to predict coastal waste risk using global plastic waste production data and related environmental and socioeconomic indicators. Seven classification algorithms, namely Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, Support Vector Machine (SVM), k-Nearest Neighbors (KNN) and CatBoost, were evaluated to distinguish between high and low risk coastal areas. Data preprocessing, feature engineering, exploratory data analysis, and cross validation were incorporated into the modeling process. The results show that Gradient Boosting achieved the highest reported accuracy of 89%, followed by Random Forest at 87%. CatBoost achieved an accuracy of 81%, while Decision Tree, SVM, KNN, and Logistic Regression achieved accuracies of 79%, 78%, 77% and 75%, respectively. The analysis also identified mismanaged plastic waste, coastal population density, river discharge and proximity to ports as important factors associated with coastal waste risk. These findings suggest that ensemble learning methods can support data driven coastal pollution assessment and help identify areas that may require priority attention. The proposed approach has potential applications in environmental monitoring, waste management planning, and resource allocation. Further research using high resolution geographical data and field validated coastal waste measurements is needed to assess its reliability and applicability across different coastal regions.