In the realm of information-driven applications crucial for emergency dispatch, activity recommendation, and news forecasting, instantaneously identifying local events such as crimes, forest fires, earthquakes, and disasters is paramount. This project addresses the challenge of efficiently classifying tweets from one of the most pivotal communication hubs, Twitter, into emergency and non-emergency categories, preceding the imperative geo-tagging.
Social media, particularly Twitter, has become a swift source of news compared to traditional channels or newspapers. Leveraging geo-tagged tweet data, we aim to develop robust machine learning models capable of capturing nuanced, metaphorical meanings within human-written text. The project delves into machine learning principles and classification algorithms, including Naïve Bayes, K-Nearest Neighbour (KNN), Decision Trees, and Random Forest.
The dataset used for this project is sourced from Twitter, currently comprising X tweets. This wealth of information on diverse social media platforms has revolutionized emergency dispatch and rescue processes, overcoming previous impracticalities.
This study incorporates potent text pre-processing methods preceding text classification, exploring the effectiveness of various machine learning models on extensive textual data. The models include Naïve Bayes and its variations, KNN, Decision Trees and Random Forest.
The project provides insights into the evolving landscape of real-time event detection and information comprehension, focusing on disaster, emergency services, machine learning, Random Forest, Naïve Bayes, Decision trees and KNN classifier.
Note
Disclaimer: The dataset used in this project may contain text that could be considered profane, vulgar or offensive.