SAGE-XGBoost: Spatially Augmented Graph Embeddings--Machine Learning Framework for Natural Hazards Susceptibility Mapping under Data Scarcity

SAGE-XGBoost introduces SAGE, a feature-engineering framework designed for natural hazards susceptibility mapping during data scarcity. To overcome limited labeled data and improve generalizability, SAGE combines controlled noise-based data augmentation with neighborhood-based graph embeddings. The process involves constructing a K-nearest neighbor graph to derive local spatial statistics. These statistics are then reduced via principal component analysis and integrated with spatial coordinates and environmental covariates to enhance prediction accuracy where conventional machine learning and deep learning models typically struggle.