A Disaster Response Pipeline

This disaster response pipeline classifies disaster messages collected by Figure Eight.

The pipeline consists of three parts:

1. ETL pipeline: a data cleaning pipeline
2. ML pipeline: a machine learning pipeline
3. Flask Web App: a data visualisation web app

ETL pipeline

The ETL pipeline loads and cleans the original datasets provided by Figure Eight. The final output is a single dataframe stored in an SQL database. All necessary files are stored in the data folder.

• categories.csv
• messages.csv
• DisasterResponse.db
• process_data.py
• ETL Pipeline Preparation.ipynb

In order to run the ETL pipeline enter the following command into your terminal: python data/process_data.py data/messages.csv data/categories.csv data/DisasterResponse.db.

ML pipeline

The ML pipeline loads the data stored at the end of the ETL pipeline. It then trains and tests the ML model. In the final step the ML pipeline stores the ML model as a pickle file.

The ML model consists of the following parts:

• CountVectorizer with a tokenizer that prepares the messages for adequate NLP
• TfidfTransformer
• SGDClassifier, a Linear Support Vector Machine

In the models folder, the following files are provided:

• classifier.pkl
• train-classifier.py
• ML Pipeline Preparation.ipynb

In order to run the ML pipeline enter the following command into your terminal: python models/train_classifier.py data/DisasterResponse.db models/classifier.pkl.

Flask App

The Flask App is a browser based application. It can classify new messages with the trained model. Furthermore, two plots based on the training messages data are provided. The first one shows the distribution of the messages' genre and the second one depicts their label distribution.

In order to run the Flask App enter the following command into your terminal: python app/run.py.