-
Customer Churn prediction is an important factor for business success, and is the focus of this project. The current library covers different steps to succeed on this task, including: pre-process data, training a ML model, predict labels for unknown data, evaluation, and model interpretability.
-
This project is part of the ML DevOps Engineer Nanodegree (Udacity).
In this work it is used a credit card customer dataset from Kaggle (https://www.kaggle.com/datasets/sakshigoyal7/credit-card-customers). The dataset consist of 10000 customers with a set of 21 demographic features, such as age, salary, marital_status, gender, credit limit, etc. This dataset is highly imbalanced, with only 16% of customer churn.
The next table provides a description of the customer variables:
| Feature | Description |
|---|---|
| CLIENTNUM | Unique identifier for the customer holding the account |
| Attrition_Flag | If the account is closed then Attrited Customer else Existing Customer |
| Customer_Age | Customer's Age in Years |
| Gender | M=Male, F=Female |
| Dependent_count | Number of dependents |
| Education_Level | Educational qualification of the account holder |
| Marital_Status | Married, Single, Divorced, Unknown |
| Income_Category | Annual income category of the account holder |
| Card_Category | Type of Card (Blue, Silver, Gold, Platinum) |
| Months_on_book | Period of relationship with bank |
| Total_Relationship_Count | Total number of products held by the customer |
| Months_Inactive_12_mon | Number of months inactive in the last 12 months |
| Contacts_Count_12_mon | Number of contacts in the last 12 months |
| Credit_Limit | Credit limit on the credit card |
| Total_Revolving_Bal | Total revolving balance on the credit card |
| Avg_Open_To_Buy | Open to buy credit line (Average of last 12 months) |
| Total_Amt_Chng_Q4_Q1 | Change in transaction amount (Q4 over Q1) |
| Total_Trans_Amt | Total transaction amount (Last 12 months) |
| Total_Trans_Ct | Total transaction count (Last 12 months) |
| Total_Ct_Chng_Q4_Q1 | Change in transaction count (Q4 over Q1) |
| Avg_Utilization_Ratio | Average card utilization ratio |
Note: variable descriptions were taken from https://ceur-ws.org/Vol-3026/paper17.pdf.
The Attrition_Flag tell us if a customer churns or not. In other words, this is the response variable to predict.
The library includes the following steps:
a) Import data
b) Exploratory data analysis
c) Feature engineering
d) Training models (Random forest, and Logistic regression)
e) Evaluation report
The project have the next tree structure:
root/
- churn_library.py
- churn_script_logging_and_tests.py
- constant.py
data/
- bank_data.csv
| File | Description |
|---|---|
| churn_library.py | Main file to run the ML pipeline |
| churn_script_logging_and_tests.py | Perform a test run and log process for inspection |
| constant.py | Configuration parameters and hyperparameters |
| bank_data.csv | Customer churn dataset |
After running churn_library.py, new directories are created to save artifacts:
root/
images/
eda/
- <ARTIFACT_NAME>.png
results/
- <ARTIFACT_NAME>.png
models/
- <MODEL_NAME>.pkl
To inspect the process, you can make a test run (optional) with churn_script_logging_and_tests.py, and a log file is created:
root/
logs/
- churn_library.log
- Create a conda environment:
conda create --name <ENV_NAME> python=3.6
- Change to conda environment:
conda activate <ENV_NAME>
-
Move to root folder of this project.
-
Install requirements:
pip install -r requirements.txt
-
Move to root folder.
-
Run code from the terminal:
python churn_library.py
As explained in the previous section (Files and data description), new files are created after running step 2.
Note: The config and training parameters can be modify directly in constant.py
Optional: If you want to inspect and log the process, enter:
python churn_script_logging_and_tests.py
The logs can be reviewed in logs/churn_library.log file.