📊 Telco Customer Churn Prediction - MLOps Project

⚠️ Note: This project requires scikit-learn 1.7.0 and imbalanced-learn 0.14.0. Using different versions may cause compatibility issues.

📌 Project Overview

This project implements an end-to-end machine learning pipeline for predicting customer churn in a telecommunications company. By identifying customers likely to cancel their services, the company can implement targeted retention strategies.

Key Features

• Data Analysis: Exploratory data analysis with visualizations
• Machine Learning Pipeline: Modular preprocessing and model training
• Multiple Models: Comparison of models from baseline to advanced ensemble methods
• Production-Ready: FastAPI implementation for real-time inference
• Business Insights: ROI calculations and customer segmentation

🏢 Business Context

This project helps telecommunications companies:

• 🔍 Identify high-risk customers before they churn
• 🧩 Segment customers for targeted retention efforts
• 💰 Estimate the financial impact of retention campaigns
• ⚡ Deploy real-time predictions for new customer data

📁 Project Structure

Main components:

• 📓 Notebooks: Step-by-step workflow from EDA to business impact analysis
• 💻 Source Code: Modular Python code for all pipeline components
• 🤖 Models: Trained machine learning models
• 📊 Data: Raw and processed datasets
• 📈 Reports: Generated outputs and business insights

For a detailed directory structure of the project, see PROJECT_STRUCTURE.md.

🚀 Installation

1. Clone the repository

   git clone https://github.com/deaneeth/telco-churn-prediction-mlops.git
   cd telco-churn-prediction-mlops

2. Create a virtual environment and install dependencies

   python -m venv venv
   .\venv\Scripts\activate  # Windows
   # OR
   source venv/bin/activate  # macOS/Linux
   pip install -r requirements.txt

📋 Dataset

The dataset contains customer information including:

• 👤 Demographics (gender, age, partner status)
• 📝 Account details (tenure, contract type, payment method)
• 📞 Services subscribed (phone, internet, add-ons)
• 💵 Financial information (monthly charges, total charges)
• 🎯 Target variable: customer churn status

🔄 Workflow

1. 🔍 Data Exploration: Analysis of customer churn patterns and feature relationships
2. 🧹 Preprocessing: Data cleaning, feature engineering, and preprocessing pipeline
3. 🛠️ Model Development: Training multiple models and optimizing performance
4. 📊 Evaluation: Comprehensive metrics and model comparison
5. 📦 Production Pipeline: Packaging for deployment
6. 💼 Business Analysis: Customer segmentation and ROI calculations

📈 Results

Model Evaluation Results

🚀 Improvement Note: These results can be further improved by implementing:

• Advanced feature engineering techniques (RFM analysis, customer behavior patterns)
• Deep learning approaches (Neural networks for complex pattern recognition)
• Hyperparameter tuning with more extensive search space
• Better handling of class imbalance using advanced sampling techniques

This project will be enhanced with these improvements in the near future.

💡 Key Insights

• 🥇 CatBoost achieved the highest F1 score (0.627859) with the best balance of precision and recall
• 📊 LogisticRegression shows strong performance with high ROC-AUC (0.847129)
• 🎯 XGBoost provides the highest precision (0.685714) but with lower recall
• 📈 The ensemble models generally outperform the baseline models (Decision Tree)

🔑 Feature Importance

• 📝 Contract type, tenure, and service issues are the strongest predictors
• ⚠️ Month-to-month contracts with technical issues show highest churn rates
• 💰 Targeted interventions show 3-5x return on investment

🚀 Deployment

The model will be deployed as a real-time prediction service using FastAPI:

will be updated...

🔮 Future Work

• 🔄 Model monitoring and retraining pipeline
• 🗃️ Feature store for reproducibility
• 🔍 Advanced model interpretability
• 🧪 A/B testing framework for retention strategies