A LSTM–XGBoost Integrated Model for Smart Resilience Networks

Natural disasters still act as a serious threat to the world structure, economy, and human lives. Effective disaster prediction, response, and recovery rely on rapid and precise disaster prediction systems. This chapter presents DisasterGuard-AI, a new hybrid framework using AI to forecast real-time disaster resilience and recovery by incorporating long short-term memory (LSTM) networks with eXtreme Gradient Boosting (XGBoost). The proposed framework gathers multimodal information from sensors, satellite imagery, and weather data, to identify temporal distributions and nonchanging characteristics for disaster impact. The XGBoost model provides a strong classification tool with the help of which the LSTM component gathers information about the temporal relations of the sensor records and predicts the magnitude of disasters and the corresponding action plans. On a vast number of different disaster datasets, the DisasterGuard-AI model reaches a maximum accuracy of 95.68%, an F1-score of 0.96, and an adjusted area under the ROC curve (AUC-ROC) of 0.97, which are substantially better than those of machine learning (ML) and deep learning (DL) algorithms. The chapter shows the viability of having hybrid AI systems in smart disaster response networks as the frameworks exhibit high-level interpretability, real-time inference, and adaptive decision-making.