Interpretable Machine Learning for Landfill Cost Estimation Using XGBoost and SHAP

Abstract

Accurate landfill cost estimation is essential for effective infrastructure planning and sustainable waste management, yet conventional approaches often fail to capture the complex nonlinear interactions among engineering, environmental, economic, and logistical factors. This study aims to develop an interpretable machine learning model for landfill cost estimation using Extreme Gradient Boosting (XGBoost) integrated with SHapley Additive exPlanations (SHAP). The analysis was conducted using data from 31 landfill projects in Indonesia collected from authorized contractors with government approval. The dataset includes remaining landfill capacity, leachate treatment cost, waste density, landfill tax, retaining wall height, and transportation distance. Data preprocessing involved unit standardization, normalization, and logarithmic transformation, followed by model training and evaluation using cross-validation, MAE, RMSE, and MAPE metrics. The results show that XGBoost outperforms Random Forest, achieving lower prediction errors and a MAPE below 1%, indicating high predictive accuracy. Residual analysis confirms that the model is generally unbiased, while feature importance and SHAP analyses identify leachate treatment cost, landfill tax, and retaining wall height as the dominant cost drivers. The novelty of this study lies in integrating explainable artificial intelligence with landfill cost estimation, providing both accurate predictions and transparent decision support for landfill planning and policy development.