Telco Churn MLOps on AWS

Problem & context

Problem

Build and evaluate a reproducible customer-churn prediction workflow, then expose the selected model through a tested prediction API.

Intended users

Technical reviewers and teams evaluating a churn-model delivery workflow.

Role and contributions

End-to-end machine learning and MLOps implementation

Dataset / inputs

A telco churn dataset containing 7,043 customer records; the repository documents train, validation and test splits.

Methods & architecture

Comparable preprocessing pipelines were used for logistic regression and XGBoost. Experiments were tracked with MLflow; the selected XGBoost threshold was 0.37.

Implementation

1. Prepare and split the 7,043-record dataset
2. Compare logistic regression and XGBoost
3. Track experiments with MLflow
4. Package the FastAPI service in Docker
5. Run checks locally, in Docker and in GitHub Actions
6. Demonstrate a manual deployment on AWS EC2

Verified results

Metrics below are reported under the project's documented evaluation conditions.

Limitations & next steps

A future iteration could automate a secure cloud deployment and add production monitoring after its operating context is defined.

Attribution & licence

See the repository for current attribution and licence information.