Customer Churn Analysis

This project analyzes customer churn for a telecom company using Python. It includes data cleaning, preprocessing, visualization, and machine learning to predict customer churn.

Repository Structure

customer-churn-analysis/
│
├─ churn_analysis.ipynb           # Jupyter notebook with complete analysis
├─ WA_Fn-UseC_-Telco-Customer-Churn.csv  # Dataset
├─ requirements.txt               # Python dependencies

Features

• Data preprocessing:
  • Handle missing values
  • Encode categorical columns
  • Scale numerical features
• Exploratory Data Analysis (EDA):
  • Churn distribution
  • Churn vs contract type
  • Monthly charges distribution by churn
  • Correlation heatmaps
  • Churn rate by internet service and payment method
• Machine Learning Models:
  • Logistic Regression
  • Decision Tree (unconstrained and pruned)
  • Random Forest
  • XGBoost
• Model Evaluation:
  • Accuracy, Precision, Recall, F1-Score
  • Confusion matrices
  • ROC curves and AUC

Installation

1. Clone the repository:

git clone https://github.com/Maheen-Fatima-1/customer-churn-analysis.git
cd customer-churn-analysis

2. Install the required packages:

pip install -r requirements.txt

Usage

1. Open churn_analysis.ipynb in Jupyter Notebook.

2. Run each cell sequentially to:

   • Load and preprocess the dataset
   • Visualize patterns and correlations
   • Train and evaluate ML models
   • Compare models and visualize ROC curves

Requirements

• Python >= 3.8
• pandas
• numpy
• matplotlib
• seaborn
• scikit-learn
• xgboost

Insights

• Long-tenure customers and those with 1- or 2-year contracts are less likely to churn.
• Fiber optic users, electronic check payers, and streaming service subscribers are at higher risk of churn.
• Logistic Regression achieved the highest recall, making it best for identifying customers likely to churn.

Notes

• Each model was tuned with thresholds to maximize F1-score or recall.
• Dataset preprocessing includes handling missing TotalCharges values and encoding categorical variables.
• Target variable Churn is encoded as 1 for churned customers, 0 otherwise.

GitHub Repository

https://github.com/Maheen-Fatima-1/customer-churn-analysis

Author

Maheen Fatima Data Science Enthusiast