A Stacking Ensemble Approach for Postdisaster Building Damage Assessment

Automated building damage assessment has emerged as a valuable tool for rapid disaster management response. However, conventional methods of damage assessment are labor-intensive and pose significant hazards. This study presents a novel method that optimizes building damage classification by using unmanned aerial vehicle (UAV) footage and artificial intelligence techniques. The proposed method enhances the performance of automated building damage assessment by integrating an object detection model with image classification models using a stacking ensemble mechanism. Three meta learners, employing different machine-learning models, are investigated: a fully connected neural network (NN), random forest, and extreme gradient boosting (XGBoost). The findings suggest that stacking resulted in performance improvements, with F1 scores increasing by up to 10%. Additionally, we examined the use of the proposed approach in other disaster zones to improve the model’s ability to adapt to unseen domains, thereby enhancing its practical applicability in disaster situations. The results demonstrate the crucial importance of using a stacking ensemble mechanism to improve the precision of damage classification in automated building damage assessment, offering a robust and effective strategy for performance enhancement. These results present promising opportunities for future advancements in the field of building damage categorization using computer vision and artificial intelligence techniques.