The development of technology that greatly makes information easier to access and share between one person and another, the use of social media by the public is so high, that it causes social interaction in cyberspace to also increase. The rapid exchange of information through social is very beneficial for social scientists and computers. Text information shared by the public on social media can be classified based on emotional expressions whose text content is easily used to predict the accuracy of a catastrophic event. This prediction process can be done using NLP (Natural Language Processing) techniques.   The purpose of this article is to analyze tweets containing disaster-related keywords with a classification model, using various pre-processing techniques and TF-IDF (Term Frequency-Inverse Document Frequency) feature extraction methods. The tweets data used in this article is sourced from KAGGLE. Next, we compared several machine learning algorithms (SVM Classifier, MultinomialNB, LogisticR, XGBClassifier, Random Forest, DecisionTree Classifier, KNeighborsClas) and Deep Learning (Recurrent Neural Network (RNN)) based on f1 score values. The results of this accuracy test show that the utilization of the RNN algorithm has a higher accuracy value compared to other algorithms, so it is worthy of being used as a model for predicting the accuracy of disaster tweets information from twitter data.