Abstract:
Plastic waste leakage into the urban river basins has emerged as a critical environmental challenge
in Sri Lanka, threatening aquatic ecosystems, public health, and urban resilience. Rapid
urbanization, inadequate waste management, and poor drainage infrastructure have accelerated the
accumulation and transport of plastic debris into waterways. This study presents a geospatial and
machine learning approach to identify and model the key drivers of plastic waste leakage in selected
urban river basins. High-resolution geospatial datasets, including land use, population density,
slope, rainfall patterns, and proximity to waste disposal sites, were integrated with field-based
leakage observations. A Random Forest classifier was employed to predict leakage hotspots,
achieving a moderate level of accuracy. Feature importance analysis highlighted waste site
proximity, urban density, and hydrological factors as the dominant predictors. The confusion matrix
further illustrates the model's strengths in identifying high-risk zones; however, misclassification
in low-risk areas suggests potential improvement through the inclusion of additional environmental
and socio-economic variables. The findings offer actionable insights for urban planners, waste
management authorities, and policymakers to develop targeted interventions for mitigating plastic
waste in riverine environments.
