Disaster Event Classification on Twitter: A Comparison of LSTM and GRU Algorithms using Word Embedding FastText and GloVe

The increasing significance of handling issues pertaining to disasters in contemporary society necessitates prompt access to precise information. This study examines disaster-related matters by doing a comparative examination of Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) algorithms. The study utilizes FastText and GloVe word embedding methodologies to categorize tweets related to disasters. The main goals are to evaluate catastrophe-related issues and compare the LSTM and GRU algorithms, with the purpose of improving disaster detection and management. The results demonstrate that the LSTM model, when combined with FastText word embeddings, surpasses previous combinations, obtaining a remarkable accuracy of 80%. The technique demonstrates a precision of 81%, recall of 83%, and an F1 score of 82%, highlighting its strong ability to effectively classify tweets connected to disasters. The study demonstrates the effectiveness of deep learning methods in efficiently sorting, categorizing, and recognizing disaster-related data.