This project focuses on Predictive Customer Churn Analysis using Machine Learning. Customer churn - the loss of clients to competitors - is a critical metric in subscription-based businesses like telecom or streaming services.

The goal is to build an interpretable ML pipeline that:

• Predicts whether a customer is likely to churn
• Explains why a prediction was made using Explainable AI (XAI) techniques
• Deploys as a web application for real-time predictions

IBM Telco Customer Churn Dataset (WA_Fn-UseC_-Telco-Customer-Churn.csv)

• Size: 7,043 customer records
• Features: 21 features including tenure, monthly charges, contract type, services used, payment method, etc.
• Target: Binary classification — Churn (Yes/No)

• Dropped irrelevant customerID column
• Handled missing/inconsistent values in TotalCharges
• Binary encoding for boolean features (Partner, Dependents, PhoneService, etc.)
• One-hot encoding for categorical features (InternetService, Contract, PaymentMethod)

• Identified key patterns and anomalies in the data
• Analyzed churn distribution (class imbalance)
• Feature correlation analysis

• Engineered new features and transformed existing variables
• Applied SMOTE (Synthetic Minority Over-sampling Technique) to handle class imbalance

• Evaluated using: Accuracy, Precision, Recall, F1 Score, ROC-AUC
• Compared all models to identify best-performing algorithm
• Final model: Random Forest Classifier

• SHAP (SHapley Additive exPlanations):
  • Global feature importance via summary plots
  • Local per-prediction explanations via force plots
• LIME (Local Interpretable Model-agnostic Explanations):
  • Local explanations for individual customer predictions

minor-project-churn/
├── data/
│   └── WA_Fn-UseC_-Telco-Customer-Churn.csv   # IBM Telco Dataset
├── models/
│   ├── model_rfc.pkl                            # Trained Random Forest model
│   ├── mlp_model.pkl                            # Trained MLP model
│   └── explainer_rfc.bz2                       # Pre-computed SHAP explainer
├── templates/
│   └── index.html                              # Flask web app template
├── notebooks/
│   └── main.ipynb                              # Main analysis notebook
├── app.py                                      # Flask web application
├── requirements.txt                            # Python dependencies
├── Pipfile                                     # Pipenv config
└── README.md                                   # Project documentation

• Python 3.9+
• pip or pipenv

# Clone the repository
git clone <your-repo-url>
cd minor-project-churn

# Create virtual environment
python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

python app.py

Open http://127.0.0.1:5000 in your browser.

jupyter notebook notebooks/main.ipynb

• Best Model: Random Forest Classifier
• Key Churn Predictors: Contract type, Tenure, Monthly Charges, Internet Service type
• XAI Integration: SHAP force plots provide per-customer explanation; LIME provides local model transparency
• Deployment: FastAPI web app with real-time churn probability gauge and SHAP explanation

1. IBM Telco Customer Churn Dataset - Kaggle
2. SHAP: Lundberg & Lee (2017) -"A Unified Approach to Interpreting Model Predictions"
3. LIME: Ribeiro et al. (2016) -"Why Should I Trust You?"
4. Scikit-learn documentation - https://scikit-learn.org