End-to-End Telecom Customer Churn Prediction System with Machine Learning Operations and Real-Time Deployment.

Customer churn prediction is a critical problem in the telecommunications industry, where retaining existing customers is more cost-effective than acquiring new ones. This paper presents an end-to-end machine learning pipeline designed to predict customer churn using historical customer data, including behavioral patterns, service usage, and billing information. The proposed system goes beyond traditional model development by incorporating a modular pipeline architecture that includes data ingestion, preprocessing, feature engineering, model training, evaluation, and deployment. To ensure data reliability, data validation is performed using Great Expectations, which enforces schema and business constraints on incoming data. The model development process leverages XG-Boost, optimized through hyperparameter tuning using Optuna. Experiment tracking, including metrics such as recall, model parameters, and artifacts, is managed using MLflow to ensure reproducibility. The system is containerized using Docker to maintain consistency across environments, while continuous integration and deployment (CI/CD) are automated using GitHub Actions. For deployment, the solution utilizes Amazon Web Services, specifically leveraging ECS with Fargate for serverless container hosting and an Application Load Balancer for efficient traffic management. The trained model is exposed as a RESTful API using Fast-API, with an interactive user interface developed using Gradio to facilitate real-time predictions. This architecture demonstrates a scalable, production-ready machine learning system aligned with modern MLOps practices.