AWS AI Resource Cleanup
Automated detection and reclamation of idle cloud resources
- 8
- AWS services scanned
- 3
- recommendation classes
- ICANDIT
- 2026 · paper presented
“Idle cloud resources sit unused and forgotten, quietly driving up cost and widening the attack surface.”
Developers spin up EC2 instances, create EBS snapshots and add IAM users for temporary access, then forget them. Orphaned resources silently inflate the bill and leave unused access points standing.
An event-driven Boto3 system scans EC2, EBS, S3, IAM, CloudWatch and RDS against YAML-configurable age and inactivity thresholds, then reclaims what is safe to reclaim on a scheduled Lambda cron. A trained Random Forest predicts keep / review / delete per resource from utilisation metrics, dry-run mode previews every deletion before anything executes, and protected tags guard critical infrastructure from accidental deletes.
What makes it
work.
Multi-service scanning across eight AWS resource types with tag-based exclusion.
Statistical outlier detection for cost spikes alongside the classifier.
Flask dashboard for resources, recommendations and reports; PDF and CSV output for audit trails.
Runs locally via cron or deploys as a Lambda triggered by EventBridge, with SNS notifications on completion.
Built with.
- Python
- Boto3
- scikit-learn
- AWS Lambda
- EventBridge
- Flask
- SNS
Developed alongside the ICANDIT 2026 paper.