Hybrid Machine Learning for Anomaly Detection in Cloud Cost Optimization

Kata, Sri Hari Reddy (2025) Hybrid Machine Learning for Anomaly Detection in Cloud Cost Optimization. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (3MB) | Preview
Preview	PDF (Configuration Manual) Download (776kB) | Preview

Abstract

Cloud computing is flexible but wastes a lot of money, about 30% of the operational costs are going either unused or by over-allocation of resources. This research paper proposed a novel two-stage hybrid anomaly-detection pipeline that combines an unsupervised Isolation Forest Algorithm as the first stage to capture anomaly scores from unlabeled CloudWatch metrics, followed by a Stacking Ensemble classifier that synthesizes predictions from Random Forest, Gradient Boosting, and Logistic Regression. The system leverages temporal features (day, month, year) alongside resource utilization metrics such as CPU, memory, network traffic, disk I/O to achieve 76.25% accuracy with 74.45% sensitivity and 78.05% specificity, demonstrating balanced performance across normal and anomalous instances. A key contribution is the severity classification with values such as Low, Medium, High, and Critical that prioritises alerts based on financial impact and confidence scores, integrated with AWS Simple Notification Service for automated email notifications enabling immediate mitigation actions. We use accuracy, precision, recall, F1 score, sensitivity, specificity, and confusion matrix to evaluate our results. Results indicate that the hybrid model, especially the stacked ensemble model, improves accuracy of detection and reliability when comparing with individual supervised models. In future work, we will explore the integration of real time ingestion from AWS CloudWatch and temporal modeling to detect changes in cloud usage over time.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Kazmi, Aqeel 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:	31 Aug 2026 15:20
Last Modified:	31 Aug 2026 15:20
URI:	https://norma.ncirl.ie/id/eprint/9712

Actions (login required)

View Item