title	emoji	colorFrom	colorTo	sdk	app_port	tags	pinned	short_description	license
NLPDisasterTweets	🚨	red	gray	docker	8501	streamlit	false	App to classify tweets as real disasters (1) or non-disaster	mit

🚨 NLP Disaster Tweet Classifier — TF-IDF + Logistic Regression

A production-ready NLP text classification project that detects whether a tweet describes a real disaster event or not — deployed as an interactive Streamlit app.

Built as a fast, explainable classical ML baseline using TF-IDF and Logistic Regression with threshold tuning. This app classifies tweets as Disaster (1) or Non-Disaster (0).

🔗 Live Demo & Code

🤗 HuggingFace Space: [https://huggingface.co/spaces/EnYa32/NLPDisasterTweets]

💻 GitHub Repository: [https://github.com/EnYa32/NLPDisasterTweets]

📓 Kaggle Competition: [https://www.kaggle.com/code/enesyama/nlp-disastertweets]

What this project does

Many tweets contain disaster-related words but do not describe real disasters.
This project trains a classical NLP model to detect real disaster tweets. Many tweets contain disaster-related keywords (fire, flood, explosion) but are used in a metaphorical or casual context.

Goal: Build a model that distinguishes:

1 = real disaster event 0 = non-disaster tweet

Model

• Text features: TF-IDF (unigrams + bigrams)
• Classifier: Logistic Regression
• Metric: F1-score
• Threshold tuning: optimized on validation set (threshold = 0.43)
• Best validation F1-score: ~0.78

📊 EDA Visualizations

NLP Preprocessing Pipeline

Applied consistently in training and inference:

lowercase

punctuation removal

whitespace normalization

keyword + location enrichment

TF-IDF transform

Files in this repo

Place these files in the repository root:

• app.py
• final_clf.pkl (trained pipeline: TF-IDF + Logistic Regression)
• threshold.pkl (float threshold, e.g. 0.43)
• requirements.txt
• README.md

▶️ Run Locally

pip install -r requirements.txt streamlit run app.py

Notes This is a classical ML baseline (fast + strong).

The same artifacts can be reused in any deployment setting.