Predict Customer Churn

• Project Predict Customer Churn of ML DevOps Engineer Nanodegree Udacity

Project Description

In this project, we identify credit card customers that are most likely to churn. The basic problem that we aim to solve is:

How do we identify (and later intervene with) customers who are likely to churn?

The data set used on this Project was pulled from Kaggle.

The Project's code is the result of refactoring a Jupyter Notebook applying engineering best practices for implementing software (modular, documented, and tested). The package have the flexibility of being run interactively or from the command-line interface (CLI).

Dependencies

For Pyton 3.8

scikit-learn==0.24.1
shap==0.40.0
joblib==1.0.1
pandas==1.2.4
numpy==1.20.1
matplotlib==3.3.4
seaborn==0.11.2
pylint==2.7.4
autopep8==1.5.6
pytest==7.4.2
notebook==7.0.3

There are a number of ways to install the correct Python version and to create a virtual environment.

Using miniconda

• Install miniconda
• After configuring the shell to run miniconda, create the customer_churn environment:

$ conda create -n customer_churn python=3.8

Activate the environment:

$ conda activate customer_churn

Install the packages

$ pip install -r requirements_py3.8.txt

Files and data description

The Project is composed of the following files:

churn_library.py

The churn_library.py is a module of functions to find customers who are likely to churn. This module can also be executed as a script using the CLI.

All the data science tasks are performed in this module such as:

• EDA
• Feature engineering
• Model training

constants.py

In the constants.py module, all constants used along the Project are set.

helpers.py

Auxiliary functions used in churn_library module are implemented.

churn_script_logging_and_tests.py

Defines unit tests using pytest and generates ./log/churn_library.log when executed from the CLI.

test_logger.py

Defines logger used in churn_script_logging_and_tests.

Running Files

Running the churn_library script

Execute in the CLI the following command:

$ python churn_library.py

This command will perform the following tasks:

• Load data set in memory
• Perform EDA
  • routine log EDA results to stdout
  • routine produces images in ./images/eda
• Perform feature engineering to increase model performance
• train classification models
  • models tested:
    • logistic regression, and
    • random forest classifier (in a grid search)
  • save models to ./models
  • generate ROC curves for both models
    • image is saved in ./images/results
  • generate classification reports for logistic and best random forest classifier model
    • reports are saved in ./images/results
  • generate feature importances plot for best random forest model
    • image is saved in ./images/results

Running the churn_script_logging_and_test script

Execute in the CLI the following command:

$ python churn_script_logging_and_test.py

This command performs the unit tests defined in the script and will produce the log file located in ./log/churn_library.log