Identification of relevant disaster information from social media and online news feeds is essential for early warning and emergency response. This paper introduces a Disaster Response and Alert System with a hybrid Natural Processing (NLP) approach that classifies short-text messages into disaster categories. This proposed Disaster Response and Alert System incorporates a Long Short-Term Memory (LSTM) network approach for contextual and sequential analysis of texts. This approach is further enhanced through a Random Forest algorithm that uses TF-IDF for lexical validation. This proposed Disaster Response and Alert System uses Random Forest algorithms that further provide probabilistic support to improve confidence levels. To determine alert activation of disaster classes such as Earthquake, Flood, Cyclone, and Tsunami, a confidence level approach is used. This study is implemented as an interactive Streamlit web application that facilitates real-time text message prediction and analytical assessment. Analytical assessment is further evaluated through a confusion matrix approach that ensures both deep learning and machine learning approaches are effectively used to classify texts related to particular disaster classifications. This study introduces a hybrid approach to effectively identify and classify texts through hybrid models of Natural Processing.

Keywords: Disaster Prediction, Natural Language Processing, LSTM, Random Forest, Text Classification, Early Warning System, Social Media Analytics.