DisasterResponsePipelines

Required Libraries

• nltk 3.3.0
• numpy 1.15.2
• pandas 0.23.4
• scikit-learn 0.20.0
• sqlalchemy 1.2.12

Motivation

The project is based on a dataset provided by Figure Eight. This dataset contains messages which were exchanged during various disasters and the categories which each message belong to.

A model has been trained using Natural Language Processing and other machine learning techniques. Hence, we are capable of classifying a message and identifying relevant categories.

Further, this project includes a very basic web app, where an emergency worker can input a new message and get classification results in several categories (i.e. to identify various categories a particular message belongs to). The web app also display some visualizations of the dataset from Figure Eight.

Below are a few screenshots of the web app.

Files

• ETL Pipeline Preparation.ipynb: Description for workspace/data/process_data.py
• ML Pipeline Preparation.ipynb: Description for workspace/model/train_classifier.py
• workspace/data/process_data.py: A data cleaning pipeline that:
  • Loads the messages and categories datasets
  • Merges the two datasets
  • Cleans the data
  • Stores it in a SQLite database
• workspace/model/train_classifier.py: A machine learning pipeline that:
  • Loads data from the SQLite database
  • Splits the dataset into training and test sets
  • Builds a text processing and machine learning pipeline
  • Trains and tunes a model using GridSearchCV
  • Outputs results on the test set
  • Exports the final model as a pickle file

Acknowledgements

I wish to thank Figure Eight for dataset, and thank Udacity for advice and review.