A Machine Learning-Driven Approach for Automated Landfill Site Selection: An Experimental Study on Marmara Region, Turkiye

Abstract

This study introduces a novel machine learning (ML)-based framework for automated landfill site selection, applied to Turkiye's Marmara Region, a vital area experiencing rapid urbanization and industrial growth. Traditional methods, often reliant on subjective expert opinions and constrained by data complexity, are reimagined using state-of-the-art ML techniques, including Logistic Regression (LR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost). Eighteen critical criteria-spanning hydrogeological, environmental, and infrastructural factors-were integrated into the framework. XGBoost achieved superior performance, with an accuracy of 0.8671, significantly outperforming LR and RF. Interpretability was enhanced using Shapley Additive Explanations (SHAP), which identified land use/land cover, distance to airports, and distance to industrial areas as the most influential factors. The resulting high-precision landfill suitability maps (LSMs) provide decision-makers with a reliable tool for selecting optimal landfill sites. This framework not only advances the technical rigor of landfill site selection but also supports sustainable waste management by addressing environmental, economic, and public health considerations. The study exemplifies the transformative potential of ML in tackling complex geospatial challenges, setting a precedent for integrating artificial intelligence into environmental planning and policy-making.