Design and development of an intelligent emergency state reporting system for Puerto Rico: AI image classification

Abstract

Puerto Rico’s recurring emergencies generate large amounts of citizen-shared imagery that could support faster situational awareness. This study developed two EfficientNet-B0 models to classify emergency type and estimate damage severity. Model 1 achieved 94.07% best validation accuracy, while Model 2 achieved 79.57%. The results showed that dataset relevance influenced performance more strongly than architectural complexity alone.