Machine Learning-Based Models for Assessing Physical and Social Impacts Before, During and After Hurricane Michael

摘要

Multi-modal approach machine learning techniques have been used to examine Hurricane Michael’s physical sensor data of cloud cover temperature and social media data from Twitter to help stakeholders and government agencies consider the societal implications of hurricane impacts more thoroughly and understand how to plan for mitigating future storms as these disasters become more frequent. Data were obtained from Twitter and NOAA on Hurricane Michael and used to evaluate the relationship between the social sentiment and the physical data during severe weather events. Of all the classification methods employed in this study to evaluate sentiment, the naive Bayes classifier results showed the highest accuracy. Models of natural language processing have been developed to explain sentiment data. Future events prediction models have been tested to improve extreme weather events emergency management. The findings demonstrate that natural language processing and machine learning techniques, using Twitter data, are practical methods of sentiment analysis. This research carried out a social media sentiment analysis that could be used by emergency managers, government officials and decision-makers to make informed emergency response decisions.