Predictive Modeling for Disaster Event Detection using Support Vector Machine

Social media has become an essential part of many aspects of human life. Many people today spend more time on social media than in-person social interactions. Therefore, it is expected that the flow of information on social media is swift. Twitter, for example, can generate 500 million tweets a day containing posts from its users about various things. Although not all of them, it is common for a tweet to have essential information. Among these critical tweets are tweets about disasters that are very useful for stakeholders to send aid as soon as possible. The Twitter platform has provided quick and accurate responses to several disaster events, such as the Westgate Mall attack in Nairobi, Kenya, the bombing in Boston, and the earthquake in California. Given the importance of the information contained in tweets about disaster events, the classification of tweets, whether they have information about disaster events or not, is needed as an effort to early detection of disaster events. Researchers conducted modeling using the SVM algorithm, which tends to perform text classification better than other traditional classification algorithms. Researchers also compared performance by modifying the SVM algorithm’s kernel and C parameters. As a result, the SVM algorithm with sigmoid kernel parameter and C parameter with value 1 is the best-performing algorithm that produces an accuracy of 80.96%, precision of 90.01%, recall of 78.60%, and F1-Score of 83.92%.