Extracting Disaster Insights from Bangla News Using Text Categorization

From emergency alerts to humanitarian response planning, accurately identifying disaster types from news articles is becoming increasingly important. It provides critical information for timely decision-making and efficient resource allocation. Bangladesh, being highly vulnerable to natural disasters such as floods, cyclones, and earthquakes, often sees significant coverage of such events in its news media. However, the automatic classification of disaster types from Bangla news remains largely unexplored due to the complexities of the language and the scarcity of labeled datasets. In this work, we address this gap by developing a disaster type classification system using several machine learning and deep learning models. Among them, the BERT-based model achieved the highest performance, with an accuracy of$\text{9 1. 2 2 \%}$, effectively classifying disaster events into categories such as floods, wildfires, and cyclones. These findings highlight the effectiveness of transformer-based models in improving disaster monitoring systems for Bangla-language news data.