A Comparative Analysis of Machine Learning Models for Classifying Disaster-Related Tweets

Social media platforms have grown to be indispensable means of communication in the digital age of today. While governments and commercial companies keep an eye on these sites during crises, they find it difficult to effectively identify disaster-related material. This work classified tweets as disaster or non-disaster related using five machine learning algorithms: naïve-Bayes, SVM, logistic regression, random forest, and decision tree. With 83.02% accuracy, our tests showed that Logistic Regression outperformed all other models by means of better performance. This result implies that SVM is useful for automating catastrophe content recognition on social media platforms, hence improving emergency response systems by means of more effective public communication monitoring under crisis conditions. The findings provide a basis for creating social media automated catastrophe monitoring systems.