Assessing urbanisation and ecological integrity coupling in Malaysia using interpretable machine learning

Abstract

Urbanisation-environment interactions are increasingly non-linear, yet traditional Coupling Coordination Degree (CCD) frameworks often rely on static, linear assumptions that fail to capture complex feedback loops. This study proposes an integrated framework combining CRITIC-weighted CCD assessment with interpretable machine learning (Random Forest and XGBoost) to decode the co-evolution of urbanisation and ecological integrity. Taking Malaysia, a country characterised by rapid expansion and high spatial disparity—as a critical case study, we utilized multi-source remote sensing and longitudinal statistics across 16 states. Our XGBoost model (Test R² ≈ 0.87) outperformed traditional regressions, confirming significant non-linear and dynamic coupling effects. By applying SHAP (SHapley Additive exPlanations), we moved beyond mere prediction to identify the sustainability “tipping points”  within the system. Key findings reveal: (1) a distinct spatial decoupling in northern and east-coast regions; (2) a “hump-shaped” threshold effect for built-up expansion where coordination peaks before declining; and (3) the critical role of forest/water assets as non-linear ecological buffers. These results demonstrate that sustainable transitions depend more on spatial structure and asset management than on income growth alone. This interpretable AI-CCD framework provides a scalable, evidence-based toolkit for low- and middle-income countries to navigate the trade-offs between development and ecological preservation.