Land cover change dynamics prediction using artificial neural networks from SVM and RF classifications in the RAJUK DAP area of Dhaka Bangladesh

Rapid and unplanned urbanization in megacities of the Global South has accelerated land use and land cover (LULC) transformation, with profound implications for environmental sustainability. This study examines LULC change dynamics in the Detailed Area Plan (DAP) area of Dhaka, Bangladesh, which is one of the most rapidly urbanizing regions in South Asia; using multi-temporal Landsat imagery for the years 2001, 2011, and 2021. Two machine learning classifiers, Support Vector Machine (SVM) and Random Forest (RF), were independently applied to four land cover classes (built-up, water body, vegetation, and others) across multiple spectral band combinations, and their classification accuracies were systematically compared. SVM consistently outperformed RF across all three time periods, achieving overall accuracies of 99%, 91%, and 98% with kappa coefficients of 0.99, 0.88, and 0.97, respectively, compared to RF accuracies of 95%, 90%, and 97%. The SVM-derived LULC maps were subsequently used as inputs to an Artificial Neural Network (ANN) transition potential model to simulate future LULC scenarios for 2031 and 2041. The ANN model was calibrated using 2001 and 2011 imagery and validated against the 2021 classification, achieving a correctness of 95.67% and an overall kappa of 0.9429. Classification results reveal a dramatic expansion of built-up area from 395.75 km² in 2001 to 808.87 km² in 2021, accompanied by steep vegetation loss, trends that are projected to intensify under current trajectories. These findings, which should be interpreted with appropriate caution given the assumption of trend continuity, provide a scientific basis for sustainable land use policy in RAJUK DAP region.