A SYSTEMATIC ANALYSIS ON THE NATURAL CALAMITY DETECTION BASED ON THE TWITTER DATA

Abstract

The increasing frequency of natural disasters, such as earthquakes, floods, and hurricanes, highlights the importance of timely monitoring and early warning systems for effective disaster management. Social media platforms, particularly Twitter, have become valuable sources of real-time information during such events. This study explores the use of Natural Language Processing (NLP) and machine learning techniques to automatically classify tweets as either disaster-related or non-disaster content, aiming to improve the speed and accuracy of identifying potential crisis situations as they occur. In this work, multiple machine learning approaches, including Random Forest and DistilBERT, are applied to process and categorize tweet data. Random Forest, an ensemble learning method, utilizes multiple decision trees to enhance classification performance and handle high-dimensional feature spaces. DistilBERT, a distilled variant of BERT, is employed to capture contextual information and sequential dependencies within the text, efficiently processing large volumes of tweets while maintaining deep understanding of word relationships. The models are trained on a labeled dataset of tweets marked as "disaster" or "non-disaster." Preprocessing steps, such as text cleaning, tokenization, and padding, are performed to ensure consistent and uniform input data. Experimental results demonstrate that both approaches are effective for real-time disaster tweet detection, with Random Forest performing well in feature-driven classification tasks and DistilBERT excelling at capturing complex contextual nuances in longer tweets. The proposed framework has potential applications in crisis response systems, enabling rapid dissemination of alerts to authorities and supporting proactive disaster preparedness and management.