Multilabel Classification of Emergency Tweets using Machine Learning and Neural Networks

A system to predict the category of tweets during times of crisis (like - flood, cyclone, earthquake, wildfires, pandemic etc.) Multilabel classification was done. Here, we have used 24 categories. They are-

1. Request for Goods and Services
2. Request for Search And Rescue
3. Request for Information
4. Call to Action for Volunteer
5. Call To Action for Donations
6. Call To Action for Moving People
7. Report of First Party Observation
8. Report of Third Party Observation
9. Report of Weather
10. Report of Emerging Threats
11. Report of New Sub Event
12. Report of Multimedia Share
13. Report of Service Available
14. Report of Factoid
15. Report of Official,
16. Report of News
17. Report of CleanUp
18. Report of Hashtags
19. Report of OriginalEvent
20. ContextualInformation
21. Advice
22. Sentiment
23. Discussion
24. Irrelevant.

Of these 24 categories, 6 of them(written in bold) are actionable categories that needs immediate attention by the emergency response officers.

We have experimented with 5 statistical classifiers. They are-

1. Multinomial Naive Bayes
2. Support Vector Machine
3. Stochastic Gradient Descent
4. Decision Tree
5. Random Forest

We have also experimented with 1 neural network (Multilayer Perceptron)