An interpretable decision-making model for identifying socio-environmental transition intervals in rural wastewater management pathway planning

Abstract

In rural wastewater management, inappropriate pathway selection can lock settlements into over-engineered centralized facilities or underperforming decentralized systems, generating avoidable capital waste, inefficient resource use, and long-term operation burdens. This study develops an interpretable transition-interval identification framework to support cleaner production-oriented rural wastewater pathway planning. Using village-level data from 353 settlements in Guangdong Province, China, the framework combines class balancing, Random Forest classification, SHAP response mapping, rule extraction, administratively grouped cross-validation and spatial autocorrelation analysis to translate nonlinear model behavior into transparent screening rules between decentralized resource-recovery and centralized engineering pathways. Within Guangdong, five-fold administratively grouped cross-validation yielded a mean AUC of 0.766 ± 0.146. Direct application of the frozen Guangdong model to 8595 Shandong villages yielded an AUC of 0.525, whereas Shandong-local redevelopment under nested administratively grouped cross-validation produced a mean fold-specific AUC of 0.585 ± 0.144 for the best-performing candidate model. Two prominent empirical transition intervals were identified: a permanent population of 1000–1600 residents, representing the principal scale-related transition interval, and an NH 3 -N pollution intensity of 3.1-3.2 kg·person −1 ·yr −1 , representing the primary pollution-related refinement. Settlement compactness and topographic slope showed additional context-dependent associations with model-predicted pathway preference. Spatial autocorrelation analysis revealed pronounced geographic clustering of observed pathway choices (Moran's I = 0.885), with centralized engineering pathways clustering in high-density coastal plains and decentralized resource-recovery pathways prevailing in mountainous settlements. The resulting boundary system provides county-level planners with a data-derivable screening tool based on routinely available village indicators, helping reduce infrastructure-context mismatch and improve resource-efficient wastewater investment before detailed project-level design.

The authors' abstract, as published at the source. Journal of Cleaner Production, 2026 · DOI ↗