Automated slum classification using deep convolutional neural networks and K-means clustering: a comprehensive Mumbai metropolitan region analysis

Rapid urbanization has led to widespread informal settlements, creating challenges for sustainable development and urban planning. This study develops an automated slum classification system using deep Convolutional Neural Networks combined with K-means clustering to analyze high-resolution satellite imagery. The approach integrates supervised deep learning with unsupervised pattern recognition to overcome limitations in current remote sensing methods. The system processes Google Earth imagery through GDAL preprocessing, applies modified CNN architectures for pixel-level classification, and produces GIS-compatible outputs via QGIS integration. Testing on Mumbai metropolitan datasets shows strong performance: 89% accuracy in airport-adjacent areas, 0.87 precision, 0.91 recall, and a 45-minute processing time per square kilometer. The method effectively classifies five land-use types: informal settlements, formal urban areas, vegetation, water bodies, and barren land, with slum detection false positives reduced to just 8%. We paid special attention to areas near airports, where informal settlements create safety concerns. The system works with standard GIS software that urban planners already use, making it practical for real-world applications in city planning and slum upgrading programs.