Tools
Tools
Samuthira Pandian Ganesan, Brindavan (2025) Automated Efficient ML Model Retraining by MLOps Using Cloud-Native Tools. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (1MB) | Preview |
Preview |
PDF (Configuration Manual)
Download (4MB) | Preview |
Abstract
Modern machine learning systems rarely fail because of a single bad model. Instead, they drift slowly as live data and usage patterns change. The methods used by many organisations to retrain models manually are still ad hoc scripts and informal data scientist to operations team handovers. This project creates and deploys a retraining pipeline which is automated connecting data versioning, training in the cloud, CI/CD, and production monitoring into one workflow of MLOps.
The system is built around the Adult Income dataset and an XGBoost classifier. Training is orchestrated on Amazon SageMaker, with data and model artifacts tracked through DVC and S3 storage. A FastAPI inference service is deployed on AWS ECS using a Docker image produced by GitHub Actions. Prometheus and Grafana monitor both predictive performance and data drift via Population Stability Index (PSI), while a Prometheus Pushgateway bridges on-demand predictions with the scrape-based monitoring model.
Experiments compare a manual baseline with the automated pipeline. The automated approach reduces human retraining effort from roughly thirty minutes of shell and console work to a single GitHub push, while maintaining or improving classification accuracy. Drift simulations demonstrate that once PSI for selected features crosses a threshold, the system can trigger a new SageMaker training job and deploy an updated image to ECS. Overall, the pipeline shows that it is possible to build an efficient, cloud-native retraining loop with mostly open-source tools, and that this approach is aligned with current industry practices for scalable MLOps.
| Item Type: | Thesis (Masters) |
|---|---|
| Supervisors: | Name Email Heeney, Sean UNSPECIFIED |
| Subjects: | T Technology > T Technology (General) > Information Technology > Cloud computing Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning |
| Divisions: | School of Computing > Master of Science in Cloud Computing |
| Depositing User: | Ciara O'Brien |
| Date Deposited: | 01 Sep 2026 10:41 |
| Last Modified: | 01 Sep 2026 10:41 |
| URI: | https://norma.ncirl.ie/id/eprint/9733 |
Actions (login required)
 |
View Item |