Disaster Management Ecosystem Using Crowdsourced Data, Deep Learning and NLP

In recent years, the frequency and intensity of natural disasters have increased, necessitating more efficient and responsive disaster management systems. This paper presents a novel disaster management solution leveraging crowdsourced data to enhance disaster response and coordination. The system bridges the gap between citizens and emergency management officials by composing a mobile application for citizens to send real-time disaster alerts and a web application for administrators to monitor and manage these alerts. A customized DBSCAN algorithm, adapted to use haversine distance, is deployed to detect clusters of disaster alerts in real time, indicating potential disaster events. Upon cluster detection, administrators can delineate affected areas on a map, triggering notifications to users within these regions and enabling them to send help requests with images and messages. These help requests are processed using the BLIP model for image captioning and the ChatGPT API for categorization, assigning relevant tags to each request. The system provides administrators with real-time data analysis features, including map filtering and citizen help request category visualization, enabling more informed decision-making. Utilizing a cloud-based architecture ensures scalability, reliability, and efficient processing of large volumes of data. This research demonstrates the system's capability to provide timely alerts and comprehensive situational awareness, ultimately contributing to more efficient disaster management.