Implementation of GA-Based Feature Selection in the Classification and Mapping of Disaster-Related Tweets

Abstract

The extracted features from Twitter messages were transformed into feature vector matrix for which feature selection using an improved Genetic Algorithm was applied. The features selected were used to train and test the classifiers. The evaluation showed the effectiveness of the implemented feature selection method in the dimensionality reduction of the feature space and in increasing the accuracy of Multinomial Naive Bayes. Moreover, a web-based prototype utilizing the model was developed and was used to analyze tweet data pertaining to natural disasters in the Philippines. The prototype exhibited potential to harness the capability of social media as a tool in helping the affected community in times of natural crisis. This work may spark ideas for a more advanced development of IT-based disaster management applications.