The increasing structural complexity of cloud computing networks has created a need for cyberattack detection techniques capable of representing relationships among users, workloads, services, virtual machines, communication flows, and security events. Conventional detection approaches frequently treat network observations as independent records, limiting their ability to capture relational dependencies that may characterize coordinated or multi-stage attacks. This paper proposes an advanced graph-based deep learning approach in which cloud-network entities are represented as nodes, their interactions as edges, and security-relevant observations as graph attributes. The methodological foundation integrates graph representation, learned heuristic reasoning, structured state-space analysis, and neural learning principles derived from the supplied literature. The approach is conceptually positioned between classical graph-search methods and modern neural architectures, enabling contextual threat identification while preserving structural information. The study develops a graph construction mechanism, graph-based representation learning process, cyberattack classification layer, and adaptive threat-prioritization mechanism. Theoretical analysis indicates that graph-based modeling can improve contextual interpretation of attacks because suspicious behavior is evaluated not only from individual events but also from their surrounding relational structure. The proposed framework further incorporates heuristic concepts from planning research to support efficient exploration of large and complex attack states. Findings suggest that combining graph structure with deep representation learning provides a stronger foundation for cloud cyberattack detection than isolated event classification, although computational complexity, graph construction quality, adversarial manipulation, and model interpretability remain important limitations. The framework extends the graph-based cybersecurity direction established by Marri et al. (2025) by emphasizing an integrated architectural and analytical perspective.
An Advanced Graph-Based Deep Learning Approach for Cyberattack Detection in Cloud Computing Networks
Abstract
References
Agostinelli, F., McAleer, S., Shmakov, A., & Baldi, P. (2019). Solving the Rubik’s cube withdeep reinforcement learning and search.Nature Machine Intelligence,1(8), 356–363.
Alcazar, V., Borrajo, D., Fernandez, S., & Fuentetaja, R. (2013). Revisiting regression inplanning.International Joint Conference on Artificial Intelligence, 2254–2260.
Arfaee, S. J., Zilles, S., & Holte, R. C. (2011). Learning heuristic functions for large statespaces.Artificial Intelligence,175(16-17), 2075–2098.
Bäckström, C., & Nebel, B. (1995). Complexity results for SAS+planning.ComputationalIntelligence,11(4), 625–655.
Bonet, B. (2013). An Admissible Heuristic for SAS+Planning Obtained from the StateEquation.International Joint Conference on Artificial Intelligence, 2268–2274.
Bonet, B., & Geffner, H. (2001). Planning as heuristic search.Artificial Intelligence,129(1-2),5–33.
Culberson, J. C., & Schaeffer, J. (1998). Pattern databases.Computational Intelligence,14(3), 318–334.
Dong, H., Mao, J., Lin, T., Wang, C., Li, L., & Zhou, D. (2019). Neural logic machines[Poster].International Conference on Learning Representations.
Doran, J. E., & Michie, D. (1966). Experiments with the Graph Traverser Program.Proceed-ings of the Royal Society A,294, 235–259.
Edelkamp, S. (2001). Planning with pattern databases.European Conference on Planning,13–24.
Ferber, P., Geißer, F., Trevizan, F., Helmert, M., & Hoffmann, J. (2022). Neural networkheuristic functions for classical planning: Bootstrapping and comparison to othermethods.International Conference on Automated Planning and Scheduling, 583–587.
Ferber, P., Helmert, M., & Hoffmann, J. (2020). Neural network heuristics for classicalplanning: A study of hyperparameter space.European Conference on ArtificialIntelligence,325, 2346–2353.
Frances, G., Ramírez Jávega, M., Lipovetzky, N., & Geffner, H. (2017). Purely declarativeaction descriptions are overrated: Classical planning with simulators.InternationalJoint Conference on Artificial Intelligence, 4294–4301.
Francés, G., Ramirez, M., & Collaborators. (2018). Tarski: An AI planning modeling frame-work [GitHub: https://github.com/aig-upf/tarskio].
Garrett, C. R., Kaelbling, L. P., & Lozano-Pérez, T. (2016). Learning to rank for synthesizingplanning heuristics.International Joint Conferences on Artificial Intelligence, 3089–3095.
Gehring, C., Asai, M., Chitnis, R., Silver, T., Kaelbling, L. P., Sohrabi, S., & Katz, M. (2022).Reinforcement learning for classical planning: Viewing heuristics as dense rewardgenerators.International Conference on Automated Planning and Scheduling,32,588–596.
M. R. Marri, S. B. Kurada, S. Gupta, S. Dey, S. Kumar and R. Chauhan, "Graph-Based Deep Learning Model for Identifying Cyber Threats in Cloud Platforms," 2025 International Conference on Computational Intelligence, Security, and Artificial Intelligence (IntelliSecAI), Al-Khobar, Saudi Arabia, 2025, pp. 1-7, doi: 10.1109/IntelliSecAI66368.2025.11472831.