Predict Customer Churn

• Project Predict Customer Churn of ML DevOps Engineer Nanodegree Udacity

Project Description

The project is a package to predict customer churn. It provides

1. An comprehensive EDA;

2. A pipeline to train and save two ML models (Random Forest + Logistic Regression) for customer churn;

The package includes a comprehensive test suite.

Files and data description

Input data:

• Customer data available in data/bank_data.csv

Generated files from EDA:

• images/eda/Churn.png: a histogram of how many customers have churned;

• images/eda/Customer_Age.png: a histogram of the age of the churning customers;

• images/eda/heatmap.png: a heatmap of the correlation between features:

• images/eda/marital_status.png: a histogram of the marital status of the churning customers

• images/eda/Total_Trans_Ct.png: a histogram of the total transactions of the churning customers;

Generate files from execution:

• /images/results/classification_report_rf.png: the classification report of the random forest model;

• /images/results/classification_report_lr.png: the classification report of the logistic regression model;

• /images/results/feature_importances.png: a plot of feature importances for the random forest model;

• /images/results/roc_plot.png: a plot of the ROC curves for RF and LR models;

• /models/rfc_model.pkl: the trained random forest model;

• /models/logistic_model.pkl: the trained logistic regression model;

Running Files

To install the package:

poetry install

or with pip:

pip install .

To generate the data:

poetry run python mlops_churn/churn_library.py

To execute the tests:

python run pytest -v tests/churn_script_logging_and_tests.py

To create baseline for images:

pytest tests/churn_script_logging_and_tests.py --mpl-generate-path=baseline

the images created are then compared on subsequent test.

The default is executed without retraining the two models. To retrain the two models, set the value RETRAIN = True.