Enhancing Cloud Service Efficiency with Predictive AutoScaling: A Machine Learning-Based Approach

Binu, Pavan (2025) Enhancing Cloud Service Efficiency with Predictive AutoScaling: A Machine Learning-Based Approach. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (977kB) | Preview
Preview	PDF (Configuration Manual) Download (963kB) | Preview

Abstract

Efficient resource management and autoscaling are critical to the performance and cost effectiveness of large-scale cloud and cluster computing systems. It is possible to predict when a computing task will eventually be at its final state, i.e. whether to terminate successfully, fall or be preempted and as a result, allocate additional resources and schedule them proactively to increase efficiency and reliability of a system. This paper reports on a study about the use of deep learning models for this predictive task. Based on a real-world large-scale platform-derived dataset, the given paper designs, implements, and evaluates three capable types of neural networks: a Convolutional Neural Network (CNN), a Long Short-Term Memory (LSTM) network, and a deep feedforward network based on Deep Q-Network (DQN) models. The research procedure is conducted with a high level of rigour and consists of extracting data appropriately formatted in compressed archives, broad data cleaning and preprocessing, feature engineering and model training in Google Colab. Results of all the models are discussed critically based on canonical classification measures, such as accuracy, precision, recall, and F1-score. After this analysis, the data reveals that all the models are relatively and similarly accurate, but their efficacies are different in various classes of task statuses. LSTM model (76.69% Accuracy) partially outperforms the other models and has a more levelled predictive power after classes. Within this report, the full lifecycle of the research will be described including data collection, the process of data preparation, comparison of the model performances, and conclude with an understanding of how deep learning can be used to predict the state of systems and what future work can be done including the proposed model set implementation of a hybrid architecture.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Makki, Ahmed 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:	20 Mar 2026 11:51
Last Modified:	20 Mar 2026 11:51
URI:	https://norma.ncirl.ie/id/eprint/9202

Actions (login required)

View Item