• GIS and ML identify 38% of Narayanganj as suitable for agro-industries. • Proximity to roads, residential areas, and rivers most influence suitability. • Only 8.2% of suitable areas overlap with official industrial zones. • ML models Random Forest and XGBoost deliver accurate suitability results. • Study reveals need for data-driven planning to protect agricultural land. Unplanned industrialization is a major driver of agricultural land loss and environmental degradation in rapidly urbanizing peri -urban regions such as Narayanganj City Corporation. Identifying suitable locations for agro-industrial growth is essential for balancing economic expansion with ecological protection. This study enhances existing site-zoning research by integrating ensemble tree-based machine learning (Random Forest and XGBoost) with a GIS framework to quantitatively prioritize spatial determinants of agro-industrial suitability. Nine geospatial variables were assessed, including proximity to major and minor roads, rivers, waterbodies, settlements, agricultural land, watersheds, and terrain factors. Model performance was strong (RF accuracy: 93.94%, XGBoost accuracy: 93.76%; AUC > 0.94). A key novelty of this study lies in combining these ensemble models with SHAP explainability to explicitly reveal how each spatial factor drives suitability, offering transparent and interpretable insights rarely applied in agro-industrial zoning. The final suitability map shows that 38% of NCC is suitable for agro-industry, while only 13% is highly suitable. Approximately 26.7% is unsuitable, increasing to 62% when waterbodies are excluded from development consideration. Notably, only 8.2% of ML-identified suitable areas overlap with industrial zones designated in the NCC Detailed Area Plan, indicating a significant spatial and policy misalignment. The results further identify proximity to major roads, residential areas, and rivers as the most influential variables, providing planners with clear priorities for decision-making. By integrating interpretable ML with spatial policy comparison, this study offers a novel, data-driven framework for guiding environmentally responsible and context-specific agro-industrial expansion, while also highlighting institutional gaps and opportunities for improved land-use governance.
Mujtabe et al. (Sun,) studied this question.