Categorization of Earthquake-Related Tweets Using Machine Learning Approaches

The Philippines is one of the most natural hazard prone countries in the world. Social media continues to be an essential part in the lives of Filipinos who have access to the Internet and the country is named as the "social networking capital of the world." The earthquake-related tweets collected were classified and manually annotated based on the four (4) labels identified namely, drill/training, earthquake feels, extent of damages, and government measures and rehabilitation. With the standard metrics, the method which obtained the highest rating is the 15 folds validation using SVM. This obtained 82.60% precision metric, recall obtained 82.50%, F-Measure resulted to 82.50%, and the correctly classified instances is 82.48%. SVM consistently showed a high evaluation performance as compared to Naive Bayes, and Linear Logistic Regression. Also, among the methods employed, the 15 folds validation consistently obtained the highest rating as compared to the 10-folds validation. Results proves that one of the advantage of SVM is its good generalization capacity in small-size training set problem and it is said to be the most effective algorithm in solving many classification problems. The next step is to conduct an evaluation to select members of the League of Albay Disaster Association (LADA) for validation of results.