A Hybrid RF-CNN Framework for GIS-Driven Land-Use Modeling and Sustainable Urban Growth Prediction

Abstract

Rapid urbanization presents significant challenges for sustainable land-use planning in developing regions. This study proposes an integrated Geographic Information System (GIS) and machine learning framework for multi-temporal land-use classification and urban growth prediction. Utilizing Landsat and Sentinel-2 satellite imagery spanning 2000 to 2023, the proposed approach combines spectral and spatial feature extraction with a hybrid Random Forest–Convolutional Neural Network (RF-CNN) classifier. The framework achieves an overall classification accuracy of 93.4% with a Kappa coefficient of 0.91, outperforming conventional methods including Maximum Likelihood Classification (78.2%), Support Vector Machines (84.5%), and standalone Random Forest (87.1%). A cellular automata–Markov chain model integrated within the GIS environment projects future urban expansion scenarios for 2030 and 2040. The results reveal that built-up areas have increased by 72.7% over the study period, while vegetation cover has declined by 50%. The proposed framework provides urban planners with a data-driven decision support tool for formulating sustainable development strategies that balance economic growth with environmental conservation.