Python Customer Churn Analysis

An end-to-end Python analytics case study that examines customer churn patterns using data-quality checks, exploratory segmentation, visualization, and interpretable predictive modeling.

The project uses a fully synthetic dataset and contains no real customer, employer, or confidential business information.

Project Overview

Customer churn can affect recurring revenue, acquisition efficiency, and long-term customer value. This project demonstrates a reproducible workflow for:

• Assessing dataset quality
• Cleaning and validating customer records
• Comparing churn across customer segments
• Visualizing churn-rate differences
• Building an interpretable logistic-regression model
• Translating analytical results into business-focused findings

Dataset

The dataset contains 800 synthetic customer records and 14 columns covering:

• Customer tenure
• Monthly and total charges
• Contract type
• Internet service
• Payment method
• Support tickets
• Late payments
• Satisfaction score
• Automatic payment enrollment
• Paperless billing
• Senior-citizen indicator
• Churn outcome

The observed churn rate in the synthetic dataset is 20.38%.

Key Findings

• Month-to-month customers had the highest contract-level churn rate at 26.25%.
• Electronic-check customers had the highest payment-method churn rate at 30.86%.
• Strong positive model signals included month-to-month contracts, electronic-check payments, fiber-optic service, total charges, and recent support tickets.

These findings demonstrate analytical methods on synthetic data and should not be interpreted as causal evidence about real customers.

Model Results

A logistic-regression pipeline was trained with:

• Median imputation for numeric values
• Most-frequent imputation for categorical values
• Standard scaling for numeric features
• One-hot encoding for categorical features
• A stratified 75/25 train-test split
• Balanced class weights

Metric	Result
Accuracy	0.7450
Precision	0.4265
Recall	0.7073
ROC AUC	0.8237

The model is intended to demonstrate an interpretable portfolio workflow, not production deployment.

Repository Structure

python-customer-churn-analysis/
├── data/
│   ├── customer_churn.csv
│   └── README.md
├── outputs/
│   └── executive_summary.md
├── src/
│   └── analyze_churn.py
├── .gitignore
├── LICENSE
├── README.md
└── requirements.txt

Analysis Workflow

The script:

1. Loads and validates the dataset.
2. Removes duplicate customer identifiers.
3. Converts numeric fields safely.
4. Creates a column-level data-quality report.
5. Calculates churn rates by contract, internet service, and payment method.
6. Saves segment-level charts and summary tables.
7. Trains and evaluates a logistic-regression model.
8. Extracts model coefficients for interpretation.
9. Produces an executive summary.

Run Locally

Clone the repository and install the dependencies:

git clone https://github.com/JashvanthiPriyanka2910/python-customer-churn-analysis.git
cd python-customer-churn-analysis
pip install -r requirements.txt

Run the analysis:

python src/analyze_churn.py

The script creates an outputs directory containing:

• Data-quality results
• Segment-level churn tables
• Churn-rate charts
• Model metrics
• Confusion matrix
• Model feature effects
• Executive summary

Technologies

• Python
• pandas
• NumPy
• Matplotlib
• scikit-learn

Skills Demonstrated

• Data cleaning and validation
• Exploratory data analysis
• Customer segmentation
• Business KPI interpretation
• Data visualization
• Feature preprocessing
• Classification modeling
• Model evaluation
• Reproducible analysis workflows
• Executive communication

Limitations

• The dataset is synthetic.
• The analysis shows association, not causation.
• The model has not been externally validated.
• No production monitoring, deployment, or fairness assessment is included.
• Results should not be used to make decisions about real individuals.

Responsible Use

This repository is a portfolio demonstration. It contains no employer-confidential data, internal system information, real customer records, or personally identifiable information.

Author

Jashvanthi Priyanka

LinkedIn