Effective Tweets Classification for Disaster Crisis based on Ensemble of Classifiers

Abstract

AbstractIn the field of disaster crisis management, the utilization of social media platforms has gained significant recognition. It helps in disseminating and gathering information during disasters, offering real-time updates on events, infrastructure damage reports, and casualty information. However, the information comes with a substantial amount of irrelevant content. Some researchers have utilized machine learning classifiers for classification, which has become ineffective. Thus, this study proposes an ensemble-based approach to disaster tweet classification, using a wide array of linguistic and word embedding features, Additionally, we investigate various supervised learning algorithms and ensemble classifiers for resolving this issue. Our findings reveal that the ensemble feature sets, specifically, the fusion of TF-IDF and word embeddings, when coupled with Bagging, achieved a classification accuracy of 98.92%. This research highlights the potential of leveraging machine learning and ensemble techniques on disaster tweet classification, providing insights for improving real-time disaster response efforts.