Multi-Step Computational Graph Prediction for Cloud Workflows: A Comparative Study of Common Machine Learning and Deep Learning Methods

Ruimin Ma, Min Chengspbu, Wumuti Sairikejiang, Ovanes Petrosian

Keywords: Cloud computing; Workflow Prediction; Computational graph; Joint Multi-step Prediction

Abstract: This paper explores the application of traditional machine learning models and deep learning models in a cloud computing environment. For the first time, it advances cloud computing by performing joint multi-step prediction of tasks and their subsequent tasks within cloud workflows. We evaluated the performance of six benchmark models (LR, SVM, XGBoost, LightGBM, CNN, and GCN) in multi-step prediction tasks. The experimental results indicate that each model has its own strengths at different prediction lengths. LR and SVM models perform well across all prediction lengths, making them suitable for tasks requiring stability and consistent performance. XGBoost and LightGBM models excel in accuracy, making them ideal for tasks demanding high accuracy. Although CNN and GCN models exhibit significant fluctuations in performance across different prediction lengths, they have notable advantages in handling complex data structures and capturing the intricate relationships between tasks. In the future, we will explore more deep learning models suitable for cloud workflow prediction tasks and apply these models in fields such as finance, healthcare, and the Internet of Things to verify their effectiveness and feasibility in various application scenarios.

Submission Number: 61