Open Access

ARTICLE

Adaptive Pareto-Based Multi-Agent Decision Model for Resource Management in Cloud Business Intelligence Systems

Khamza Eshankulov¹, Bahodir Muminov², Robiya Farmonova¹, Dilnavoz Sodikova³, Bakhriddin Bozorov⁴, Zavqiddin Temirov⁵, Rashid Nasimov^6,*

1 Department of Applied Mathematics and Programming Technologies, Bukhara State University, Bukhara, Uzbekistan
2 Department of Artificial Intelligence, Tashkent State University of Economics, Tashkent, Uzbekistan
3 Department of Biomedical Engineering, Biophysics and Informatics, Bukhara State Medical Institute, Bukhara, Uzbekistan
4 Chief Innovation Officer, “Navoi Mining and Metallurgical Company” Joint-Stock Company, Navoi, Uzbekistan
5 Department of Digital Technologies, Alfraganus University, Yukori Karakamish Street 2a, Tashkent, Uzbekistan
6 Department of Computer Engineering, Gachon University, Seongnam-Si, Republic of Korea

* Corresponding Author: Rashid Nasimov. Email:

Computers, Materials & Continua 2026, 88(3), 94 https://doi.org/10.32604/cmc.2026.081803

Received 15 March 2026; Accepted 05 June 2026; Issue published 23 July 2026

Abstract

Cloud-based Business Intelligence (BI) systems operate under highly dynamic analytical workloads, including bursty OLAP queries, concurrent aggregations, and real-time microservice interactions, where static resource allocation leads to latency spikes and inefficient resource utilization. This paper proposes a decentralized adaptive Pareto-based multi-agent decision model for real-time resource coordination in cloud BI microservice environments. The agent placement problem is formulated as a multi-criteria decision process that minimizes service response latency, improves computational resource utilization, and preserves Quality-of-Service (QoS) stability. Instead of constructing a centralized global optimization policy, the proposed framework relies on decentralized locally Pareto-efficient decisions combined with adaptive priority regulation driven by QoS deviation. The approach is evaluated through large-scale controlled simulation and validated in a Kubernetes-based pilot cloud environment. Experimental results demonstrate up to 54% latency reduction compared to static allocation and 22% improvement over GA-based optimization, with enhanced CPU utilization balance under dynamic workloads. Statistical analysis confirms the significance of improvements (p < 0.05). The proposed model ensures bounded monotonic decision transitions without centralized orchestration or predictive training, making it suitable for real-time cloud-native BI service ecosystems.

---

Keywords

---