Integrating Socio-Political Dynamics and Geospatial Machine Learning: An Integrated-Hybrid ANN approach for sustainable Urban Planning in Post-Conflict Baghdad

Rapid, post-conflict urbanization in Baghdad presents acute socio-environmental and infrastructure challenges that conventional remote-sensing models struggle to capture. This study develops a hybrid geospatial–socio-political framework that integrates high-resolution Landsat/Sentinel imagery and spatial indicators (Normalized Difference Vegetation Index (NDVI), Normalized Difference Built-up Index (NDBI), Nighttime Lights (NTL), Population Density, Digital Elevation Model (DEM), Distance-to-Road/Water, Building Footprints) with socio-political rasters (UNHCR displacement statistics; subnational governance indices from the Global Data Lab) to forecast land-use/land-cover (LULC) to 2050. A multi-layer perceptron Artificial neural networks (ANN) (input = 9 predictors; hidden layers = 64–128–64; ReLU + dropout 0.3; softmax output) was trained on stratified samples (≈50,000 pixels/city) and implemented in Keras. Historical analysis (1990–2020) shows Baghdad’s built-up area rose ≈82% with mean NDVI declining ≈40%, while Riyadh’s built-up rose ≈55% with NDVI declining ≈20%. The ANN achieved ~88% overall accuracy and a Kappa of 0.82 on the test set. Projections to 2050 (medium-trend scenario) indicate further built-up increases of ≈25% for Baghdad and ≈15% for Riyadh. Feature-importance and ablation tests attribute the largest predictive contribution to displacement density (permutation accuracy drop ≈10.4%), followed by NDVI (≈8.8%) and governance indices (≈7.2%). Scenario-based sensitivity (±25% socio-political perturbations) alters Baghdad’s projected built-up share by ≈8 percentage points, underscoring high socio-political sensitivity; input extrapolation and sensor inter-calibration introduce additional uncertainty (assessed at ~±15–25% across inputs). The results argue for policy responses combining slum-upgrading, adaptive zoning, institutional strengthening, and real-time monitoring (IoT/NTL integration). Future work should apply explainable-AI methods, finer-scale socio-political data, and dynamic (feedback) models to improve causal interpretation and scenario planning.