Disaster Response Pipeline Project

Project Overview

In the wake of natural disasters, emergency-response organizations are inundated with tweets, text messages, and service‑desk tickets.
Manually triaging these requests costs precious minutes that could save lives.

This repo delivers an end‑to‑end pipeline—from raw CSV files to a production‑ready Flask app—that automatically classifies each incoming message into 36 humanitarian‑response categories (e.g., water, search and rescue, medical help), ensuring they reach the right teams fast.

The system combines standard NLP preprocessing (tokenization, lemmatization, TF‑IDF) with a grid‑searched multi‑output RandomForest/AdaBoost ensemble.
Performance is tracked with accuracy, precision, recall, and F1, and the trained model is exposed through a simple web UI so field operators can paste a message and instantly see which relief teams should respond.

---

Technical Architecture

Raw CSV ─► ETL Pipeline ─► SQLite DB ─► ML Pipeline ─► Pickled Model ─► Flask App

Interactive Jupyter Notebooks

Notebook	Purpose
ETL_Pipeline_Preparation.ipynb	Step‑by‑step exploration of the data‑cleaning workflow implemented in process_data.py.
ML_Pipeline_Preparation.ipynb	Interactive training, tuning, and evaluation of the multi‑output classifier (mirrors train_classifier.py).

Tip : No Jupyter? GitHub renders notebooks automatically.

---

1  ETL Pipeline (Extract → Transform → Load)

process_data.py:

• Reads message and category CSVs.
• Merges datasets, fixes inconsistencies, removes duplicates.
• One‑hot‑encodes the 36 category columns.
• Saves the clean result to DisasterResponse.db (SQLite).

---

2  Machine‑Learning Pipeline

train_classifier.py:

• Splits data, builds a Pipeline (TF‑IDF → classifier).
• Runs GridSearchCV over key hyper‑parameters (n_estimators, max_depth, etc.).
• Prints classification report per label and saves the best model to classifier.pkl.

Test‑set highlights:

• Accuracy : 94.4 %
• Weighted F1 : 0.86
• Solid recall on frequent labels (related, request); room for improvement on rare ones (refugees, shops).

---

Getting Started

1. Clone the repo

   git clone https://github.com/Soriano-R/disaster-response-pipeline.git
   cd disaster-response-pipeline

2. Install dependencies

   pip install -r requirements.txt

3. Build the database

   python data/process_data.py data/disaster_messages.csv data/disaster_categories.csv data/DisasterResponse.db

4. Train the model

   python models/train_classifier.py data/DisasterResponse.db models/classifier.pkl

5. Run the web app

   cd app
   python run.py

   Open http://localhost:3001 in your browser.

---

Repository Structure

├── app/
│   ├── run.py
│   └── templates/
│       ├── master.html
│       └── go.html
│
├── data/
│   ├── disaster_messages.csv
│   ├── disaster_categories.csv
│   ├── process_data.py
│   └── DisasterResponse.db         ← generated
│
├── models/
│   ├── train_classifier.py
│   └── classifier.pkl              ← generated
│
├── resources/
│   ├── Disaster_Response_Application_Interface.png
│   ├── Disaster_Response_Classification_Result.png
│   ├── ETL_Pipeline_Preparation.ipynb
│   ├── ML_Pipeline_Preparation.ipynb
│   ├── ETL_Pipeline_Preparation.html
│   └── ML_Pipeline_Preparation.html
│
├── requirements.txt
└── README.md

---

License

This project is released under the MIT License. See LICENSE for details.