Multimodal-Disaster-Event-Identification-from-Social-Media-Posts

This skill uses multimodal machine learning to classify disaster events from paired social media images and tweets. It processes visual and textual data simultaneously through separate feature extraction paths—EfficientNetB3 for image analysis and BERT for tweet processing—then fuses the results for unified classification. The model achieves 94.18% accuracy across six disaster-related categories including building damage, infrastructure damage, human casualties, and non-damage cases. It handles imbalanced datasets through preprocessing techniques including image scaling, normalization, and augmentation. Organizations responding to disasters can leverage this skill to automatically filter and categorize social media reports for rapid situational awareness.