🌪️ ResQTweet

AI-powered Real-Time Disaster Tweet Classification & Alerting System

---

🌍 Sustainable Development Goals (SDGs)

• 🌆 SDG 11 – Sustainable Cities and Communities: Safer, more resilient cities with disaster-preparedness tools.
• 🌍 SDG 13 – Climate Action: Timely action against climate-related disasters through early detection.

---

🎯 Problem

During natural disasters or humanitarian crises, timely information is critical.

• Twitter (X) provides real-time updates but is often noisy or metaphorical.
• Distinguishing real disaster tweets from irrelevant chatter is difficult.
  Challenge: Automatically detect genuine disaster-related tweets to assist responders, NGOs, and media in rapid situational awareness.

---

✅ Solution

ResQTweet leverages NLP + ML + summarization to:

• Preprocess and clean tweets.
• Extract TF-IDF features from text and keywords.
• Classify tweets as disaster or non-disaster using ML models.
• Summarize disaster tweets with HuggingFace DistilBART for concise alerts.

---

✨ Key Features

• Real-time Detection: Classifies tweets instantly with a lightweight ML model.
• Alert Generation: Produces human-readable, concise alerts.
• Keyword + Semantic Understanding: Captures context beyond simple keyword matches.
• Scalable & Extensible: Future multilingual support for global crises.

---

⚙️ Tech Stack

Layer	Tools & Frameworks
Programming	Python
Data Handling	Pandas, NumPy
ML Models	Scikit-learn (Logistic Regression, Naïve Bayes)
NLP	Hugging Face Transformers (DistilBART)
Vectorization	TF-IDF (unigrams + bigrams)
Visualization	Matplotlib, Seaborn
Deployment (Future)	Streamlit / Dash + Twitter API
Environment	Google Colab

• Dataset: Natural Language Processing with Disaster Tweets

---

📊 Model Details & Performance

Algorithms Evaluated: Multinomial Naïve Bayes, Complement Naïve Bayes, Logistic Regression

Features: TF-IDF vectors (text + keyword) + binary keyword flag

Model	Accuracy	Precision	Recall	F1-score
MultinomialNB	81.29%	0.816	0.728	0.770
ComplementNB	80.70%	0.785	0.758	0.771
Logistic Regression	81.48%	0.804	0.752	0.777

📌 Best model → Logistic Regression (balanced accuracy & F1).

---

🔄 Process Pipeline

              ┌─────────────────────────────┐
              │   Load tweets_data.csv      │
              │                             │
              └─────────────┬───────────────┘
                            │
                 ┌──────────▼─────────────┐
                 │     Preprocessing      │
                 │    - Clean text &      |
                 |       keyword columns  │
                 │    - Remove URLs,      |
                 |      mentions,hashtags,|
                 |   non-alphabetic chars │
                 │  - Lowercase all text  │
                 │  - Remove extra spaces │
                 │  - Handle NaNs         │
                 │  - Drop irrelevant cols│
                 └──────────┬─────────────┘
                            │
          ┌─────────────────┴──────────────────┐
          │                                    │
 ┌────────▼─────────┐                 ┌────────▼─────────┐
 │  TF-IDF on text  │                 │ TF-IDF on keyword│
 │(unigrams+bigrams)│                 │  (unigrams, all) │
 └────────┬─────────┘                 └────────┬─────────┘
          │                                    │
          └────────────────┬───────────────────┘
                           ▼
      ┌───────────────────────────────────────────────┐
      │          Combine Features (hstack)            │
      │ [text TF-IDF + keyword TF-IDF + keyword_flag] │
      └───────────────────┬───────────────────────────┘
                          │
             ┌────────────▼───────────────┐
             │   Split train/validation   │
             │   (stratified 80-20 split) │
             └───────────┬────────────────┘
                         │
             ┌───────────▼───────────────┐
             │        Train Model        │
             │     LogisticRegression    │
             │       MultinomialNB       |
             |       ComplementNB        │
             └───────────┬───────────────┘
                         │
             ┌───────────▼───────────────┐
             │   Evaluate Model          │
             │ Accuracy / F1-score /     │
             │ Confusion Matrix          │
             └───────────┬───────────────┘
                         │
             ┌───────────▼───────────────┐
             │   Save / Select Best Model│
             │ LogisticRegression chosen │
             └───────────┬───────────────┘
                         │
             ┌───────────▼───────────────┐
             │   Predict on New Tweets   │
             │   (raw text input)        │
             └───────────┬───────────────┘
                         │
             ┌───────────▼───────────────┐
             │  Generate Alerts          │
             │  - Summarize if disaster  │
             │  - Return 🚨 or ✅       │
             └───────────────────────────┘

---

🚨 Sample Outputs

disaster_alert_system("Earthquake of 6.1 magnitude hits California.")
# 🚨 DISASTER ALERT: Earthquake of 6.1 magnitude hits California.

disaster_alert_system("Huge flood of messages in my inbox 😂")
# ✅ No disaster detected.

disaster_alert_system("Explosion at chemical plant in Houston, authorities evacuating nearby areas.")
# 🚨 DISASTER ALERT: Explosion at chemical plant in Houston, authorities evacuating nearby areas.

---

📂 Project Structure

ResQTweet/
│── Code/                  # Code (.ipynb) & Dataset (tweets_data.csv)
│   ├── ResQTweet.ipynb
│   ├── tweets_data.csv
│── WriteUp/                   # Core scripts
│   ├── WriteUp.md
│   ├── Word Documet
│
│── requirements.txt       # Dependencies
│── README.md              # Project documentation

---

🌟 Impact

🏛️ Governments: Faster disaster response & prioritization

🆘 NGOs: Smarter humanitarian aid allocation

📰 Media: Early verification of breaking crises

🌎 Future: Multilingual expansion for global early-warning system

---

👩‍💻 Author

Vaibhavi Srivastava

🔗 Github : archangel2006

📅 Hackathon : Global AI Buildathon 2025