Anomaly detection and early risk identification in digital disaster response-based on deep learning in public health

Chun Pan^*

• Nanjing University of Information Science and Technology, Nanjing, China

In the evolving landscape of disaster response, integrating advanced digital technologies is critical to enhancing the efficiency and effectiveness of public health systems. Traditional anomaly detection methods often fall short due to their inability to handle the dynamic, heterogeneous, and real-time nature of disaster-related data. These methods typically rely on static models that struggle with integrating continuous data streams from diverse sources like hospitals, emergency services, social media, and environmental sensors. As a result, they often fail to capture sudden shifts in disease patterns, environmental conditions, or population movements, leading to delayed risk identification and suboptimal decisions. The increasing frequency and complexity of natural disasters and pandemics underscore the need for flexible, adaptive systems capable of learning from evolving data. Recent advances in machine learning, artificial intelligence, and big data analytics offer promising tools to address these limitations by enabling real-time, high-dimensional data analysis. In recent years, the integration of advanced digital technologies has become essential for improving public health disaster response. This study proposes a deep learning-based framework for anomaly detection and early risk identification during digital disaster response scenarios, leveraging data from hospitals, emergency services, social media, and environmental sensors. The objective of the study is to enhance real-time decision-making and situational awareness in public health crises. Experimental results across multiple datasets (EM-DAT, FEMA, UNOSAT, Earthquake) demonstrate that our proposed model improves anomaly detection performance by 23% in precision and reduces false alarms by 31% compared to baseline models. The method combines LSTM and transformer-based architectures to effectively analyze spatiotemporal data, offering both high accuracy and interpretability for public health experts.