Explainable AI with spatio-temporal transition modelling for predicting urban land use transitions in the Kolkata metropolitan area

Rapid, unregulated urban expansion in the Kolkata Metropolitan Area has triggered profound land-use and land-cover transformations with significant ecological and governance consequences. This study presents an interpretable modelling framework integrating ensemble machine learning, SHapley Additive exPlanations, and a Cellular Automata–Artificial Neural Network simulation to reconstruct land use dynamics from 1991 to 2025 and project transitions to 2050. Sixteen explanatory variables were assessed using Random Forest, Support Vector Machine, Extreme Gradient Boosting, and Light Gradient Boosting Machine classifiers, with SHAP-derived weights embedded as structural inputs within the simulation framework. Validated at 93% overall accuracy, projections indicate that built-up land will reach 1,146 km² by 2050, with vegetation declining by 76% and agricultural land by 46%. Accessibility drivers dominated urban transition probability, while ecological and regulatory constraints accounted for less than 8%, underscoring a persistent governance gap. A consolidated Policy Summary Map provides direct planning support for metropolitan stakeholders.