#Disaster Detection via Twitter
This repository contains a predictive classification model I developed for determining if a Tweet is discussing a disaster event or not (i.e., building collapse, wildfire, terrorist attack) as a capstone project for the Data Science course offered at General Assembly in Austin, TX. The course was excellent and I highly recommend it to anyone interested. Later at The Data Incubator I further developed this model and built an online app interface for it which can be found (here).
Form: The project is in a long python notebook (disaster_detection_via_twitter.ipynb) which can be viewed directly via the GitHub website. Alternatively, it can be viewed via the online Jupyter notebook viewer (click here). For brevity, feel free to just read the descriptions and ignore the long code segments. There is a total of eight sections with figures and plots interspersed throughout.
Outcome: combining various Natural Language Processing tools, an accurate Logistic Regression classifier is built to automatically identify a Tweet as a disaster or not. The model boasts an 8-fold cross validation accuracy of roughly 88% and a receiver operating characteristic (ROC) area of roughly 0.94.
This tool may be freely used and improved upon (in fact this is encouraged). If any improvements on the model are made please let me know.