<p>Disasters can strike unexpectedly and leave a trail of destruction, causing immense suffering and loss of life while disrupting entire communities. These events can be natural, such as floods, earthquakes, hurricanes, wildfires, or man-made, including industrial accidents and technological failures. This study investigates a hybrid approach that uses text analysis, natural language processing, and optimization techniques to identify and monitor disaster-related events. The methodology of this paper involves collecting and analyzing text, focusing on sentiment and keywords associated with disaster-related text. Various aspects of text patterns are examined to enhance the model’s performance. The proposed model uses a Levy flight-based Particle Swarm Optimization algorithm to select optimal features from a vector set. It uses Text Blob for sentiment analysis, cosine similarity to classify each tweet as a disaster, Count Vectorizer for feature extraction, and XGBoost machine learning algorithm for classification. The significance of this model is that it provides early warning and insight for any disaster based on text analysis and classification.</p>