A HYBRID LEARNING FRAMEWORK FOR POST-DISASTER DAMAGE ASSESSMENT AND SHELTER DECISION-MAKING: INCORPORATING 3D-PRINTED MODULAR ARCHITECTURE IN THE KAHRAMANMARAŞ, TÜRKİYE CONTEXT

The growing frequency and severity of natural disasters have highlighted the urgent need for adaptive, efficient, and sustainable temporary housing strategies. This study introduces a hybrid computational framework that integrates parametric design, Bayesian networks, fuzzy logic, and weakly supervised learning to enhance post-disaster temporary housing decisions. Using high-resolution aerial imagery from the 2023 Türkiye Earthquake dataset, the system extracts multi-layered spatial and structural features to classify damage levels and inform shelter typology. In addition to damage assessment and decision support, the framework incorporates fabrication-aware modules for 3D-printed modular architecture, enabling rapid, locally manufacturable shelter components tailored to site-specific needs. This integration improves deployment speed, supports modular adaptability, and aligns with Industry 4.0 principles for automated construction. The proposed SEHRNet-based architecture combines deep learning with probabilistic graphical models to accommodate both quantitative and qualitative uncertainty. A hybrid decision-making mechanism integrating TOPSIS, PROMETHEE, and Simulated Annealing enables evaluation of shelter alternatives under multiple constraints such as cost, modularity, climate compatibility, and cultural adaptability. A feedback loop based on Multi-Time-Step Rolling with MPC allows for real-time updates and adaptive planning. The results demonstrate improved decision accuracy and provide a fabrication-aware, computationally scalable solution for disaster-responsive shelter planning.