Paper Title
EVALUATING DEEP LEARNING AND TREE-BASED MODELS FOR GOVERNANCE-ORIENTED CLOUD LOAD PREDICTION USING GOOGLE CLUSTER TRACES

Abstract
Accurate prediction of cloud resources is critical to the success of e-Governance because of the need for efficient use of resources while also providing transparency, predictability, and reliability. There is a growing body of research supporting the use of deep learning models, such as LSTM, GRU and TCN, for predicting workloads. However, the use of these models in governance-based cloud environments has not been thoroughly explored. This paper aims to provide a comprehensive comparison of all deep learning architectures (ANN, LSTM, GRU, TCN) and compares them to existing non-neural ensemble models (XGBoost, LightGBM, CatBoost) on actual Google Cluster Trace data. We will evaluate the models based on prediction accuracy, training time, inference latency and memory usage — all of which are important considerations in the cloud on a public sector level. Our tests indicate that while deep learning models have more complex architectures than simpler baseline models, in many cases there is no statistical difference in terms of their predictive accuracy and there may be additional computational costs associated with using these models. Accordingly, deep learning models may not represent an appropriate approach for balancing cloud loads within the e-Government context in real time, where factors such as interpretability, stability and cost benefit outweigh the potential benefits of using deep learning models. Keywords - Cloud Computing, Load Prediction, E-Governance, Google Cluster Trace, Deep Learning, Resource Management.