Paper deep dive
Intelligent Cloud Orchestration: A Hybrid Predictive and Heuristic Framework for Cost Optimization
Heet Nagoriya, Komal Rohit
Intelligence
Status: succeeded | Model: google/gemini-3.1-flash-lite-preview | Prompt: intel-v1 | Confidence: 95%
Last extracted: 4/3/2026, 12:18:10 AM
Summary
This paper proposes a hybrid cloud orchestration framework that integrates macro-level predictive machine learning (e.g., LSTM) for long-term workload forecasting with micro-level heuristic algorithms (e.g., Game Theory) for real-time task allocation. This approach aims to balance the predictive accuracy of ML with the low-latency responsiveness of mathematical heuristics, effectively reducing cloud infrastructure costs while maintaining SLA compliance.
Entities (5)
Relation Signals (3)
Hybrid Orchestration Framework â combines â Long Short-Term Memory
confidence 98% · proposes a hybrid orchestration framework that combines LSTM-based predictive scaling with heuristic task allocation
Long Short-Term Memory â usedfor â Macro-level workload forecasting
confidence 95% · ML modules (e.g., LSTM) would be utilized strictly for macro-level, long-term workload forecasting
Game Theory â usedfor â Micro-level task allocation
confidence 95% · lightweight mathematical algorithms (e.g., Game Theory frameworks) would process this predictive intelligence to handle micro-level, immediate task-to-container allocations
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:Cloud computing allows scalable resource provisioning, but dynamic workload changes often lead to higher costs due to over-provisioning. Machine learning (ML) approaches, such as Long Short-Term Memory (LSTM) networks, are effective for predicting workload patterns at a higher level, but they can introduce delays during sudden traffic spikes. In contrast, mathematical heuristics like Game Theory provide fast and reliable scheduling decisions, but they do not account for future workload changes. To address this trade-off, this paper proposes a hybrid orchestration framework that combines LSTM-based predictive scaling with heuristic task allocation. The results show that this approach reduces infrastructure costs close to ML-based models while maintaining fast response times similar to heuristic methods. This work presents a practical approach for improving cost efficiency in cloud resource management.
Tags
Links
- Source: https://arxiv.org/abs/2604.02131v1
- Canonical: https://arxiv.org/abs/2604.02131v1
Trouble viewing inline? Open PDF directly â
Full Text
28,083 characters extracted from source content.
Expand or collapse full text
Intelligent Cloud Orchestration: A Hybrid Predictive and Heuristic Framework for Cost Optimization Heet Nagoriya Information Technology G H Patel College of Engineering and Technology Anand, India 2023002329.gcet@cvmu.edu.in Prof. Komal Rohit Information Technology G H Patel College of Engineering and Technology Anand, India komalrohit@gcet.ac.in AbstractâCloud computing allows scalable resource provisioning, but dynamic workload changes often lead to higher costs due to over-provisioning. Machine learning (ML) approaches, such as Long Short-Term Memory (LSTM) networks, are effective for predicting workload patterns at a higher level, but they can introduce delays during sudden traffic spikes. In contrast, mathematical heuristics like Game Theory provide fast and reliable scheduling decisions, but they do not account for future workload changes. To address this trade-off, this paper proposes a hybrid orchestration framework that combines LSTM-based predictive scaling with heuristic task allocation. The results show that this approach reduces infrastructure costs close to ML-based models while maintaining fast response times similar to heuristic methods. This work presents a practical approach for improving cost efficiency in cloud resource management. KeywordsâCost Optimization, Resource Allocation, Machine Learning, Cloud Computing, Task Scheduling. I. INTRODUCTION Cloud computing significantly changes IT infrastructure by providing scalable, on-demand resources. However, managing multiple workloads simultaneously introduces significant financial challenges. A primary issue is unexpected cost overruns, with companies frequently exceeding their projected cloud expenditures by up to 30%. This difference is often due to heavy over- provisioning strategies aimed at ensuring application availability and performance, which paradoxically leads to substantial financial waste instead of the intended cost savings. The core problem lies in the accumulation of underutilized storage, network, and computing resources as application complexity grows. Historically, cloud workload distribution relied on static pricing models and manual forecasting, where infrastructure was allocated based on predetermined estimations. While sufficient for monolithic applications with predictable workloads, this approach is not sufficient for modern, distributed cloud systems. Contemporary architecturesâbuilt using microservices, containerized deployments, and multi-cloud strategiesâexhibit highly dynamic workloads. Resource demand can fluctuate rapidly due to scaling events and traffic bursts. Traditional allocation models fail to respond quickly to these fluctuations, resulting in either over-provisioning (inflating costs) or under-provisioning (compromising performance and availability). As a result, modern cloud ecosystems require intelligent, autonomous resource management techniques capable of continuous workload monitoring and real-time adjustment. To address the challenges of dynamic cloud systems, this paper evaluates two main resource optimization methods: traditional mathematical optimization and data-driven Machine Learning (ML) approaches. Mathematical optimization includes techniques such as Linear Programming and Mixed-Integer Programming. These methods provide efficient resource allocation under defined constraints and help reduce financial waste. Conversely, ML uses historical telemetry data to predict future utilization patterns and dynamically adjust resource provisioning, enabling the system to adapt to load variations and unexpected traffic spikes effectively. This study shows a detailed comparison of mathematical and ML-driven methodologies and explains their respective advantages, limitations, and optimal use cases. The primary objective is to evaluate strategies that minimize cloud operational costs without degrading system latency or reliability. By analyzing both methods, this paper aims to identify pathways toward more efficient, responsive, and cost-effective cloud resource management. The main contributions of this paper are: (1) A comprehensive taxonomy categorizing recent cloud cost optimization studies into mathematical heuristics, predictive machine learning, and architectural frameworks. (2) A critical evaluation of the latency-accuracy trade-offs between resource-intensive predictive models (e.g., LSTM, Deep Reinforcement Learning) and lightweight deterministic algorithms (e.g., Game Theory, Simulated Annealing). (3) The introduction of a Hybrid Optimization-ML conceptual framework that synergizes macro-level predictive capacity scaling with micro-level, strictly bounded task allocation. I. BACKGROUND A. Definitions and cloud environments To understand cost-efficient resource allocation, it is necessary to discuss the types of cloud computing environments commonly discussed in research literature. Infrastructure as a Service(IaaS): Infrastructure as a Service or IaaS is the layer of cloud computing. Cloud computing companies offer resources like virtual machines, storage, and networking over the internet. Users typically rent these resources and only pay for what they use. This means companies can make their infrastructure bigger or smaller as they need to. When they use IaaS companies are in charge of the operating system, the software and the applications they use. Serverless Computing (Function-as-a-Service â FaaS): Serverless Computing, which is also known as Function-as-a-Service or FaaS is a way of running programs where the cloud company takes care of the servers and resources. Developers only need to write and deploy programs that do specific tasks when something happens. The cloud company makes sure everything runs smoothly and scales up or down as needed. People only pay for the time their programs are actually running so they do not have to pay for resources they are not using. Spot Instances: Spot Instances are when cloud companies offer their computing power at very low prices. This can save companies a lot of money. The catch is that the cloud company can take back the resources at any time if they need them for other customers. Because of this companies need to have plans in place like saving their work and managing their resources carefully to use spot instances effectively. Other people have already talked about how to do this, in their studies [5] [6]. B. The gap in existing literature: Recent literature extensively covers resource management, task scheduling, and performance improvements in cloud computing. While many studies address scheduling strategies, architectural optimizations, and energy efficiency [30], fewer specifically concentrate on minimizing monetary costs as the main objective. Existing research can be divided between AI/ML utilization for resource management [3][29] and lightweight scheduling algorithms [12]. Furthermore, architecture-related studies explore serverless platform economics [21] and the constraints of decentralized Edge and Fog computing environments [25]. Instead of merely describing these works in isolation, a critical comparison highlights their different techniques. For instance, optimization-based studies rely on algorithms like Simulated Annealing and Game Theory [11][33] to find immediate, near-optimal resource matches based on strict mathematical constraints. In contrast, ML-focused literature [4][20][28] uses Deep Reinforcement Learning (DRL) and Long Short-Term Memory (LSTM) networks to address the same scheduling problems by interpreting historical data and predicting future load patterns. A critical gap in this existing literature is the lack of clear comparison studies evaluating whether the computational expense and delay involved in training ML models are justified by their cost savings, especially when compared to the immediacy of deterministic mathematical optimization in volatile pricing environments like spot markets. To bridge this gap and provide a clear comparison of existing methodologies, Table I summarizes key studies, detailing their underlying algorithms, utilized datasets (or environments), achieved cost reductions, and significant limitations. Table I. Comparison of Key Cloud Cost Optimization Studies I. RESEARCH METHODOLOGY To systematically review the landscape of cost reduction techniques in cloud infrastructure, a structured literature selection process was employed. Databases Searched: Principal searches were conducted across major academic digital libraries, including IEEE Xplore, ACM Digital Library, SpringerLink, and ScienceDirect. Search Keywords: Queries were constructed using logical operators: ("Cloud Computing" OR "Serverless") AND ("Cost Optimization" OR "Resource management" OR "Task Scheduling") AND ("Machine Learning" OR "Optimization Algorithms" OR "Game Theory"). Inclusion & Exclusion Criteria: The review prioritized peer-reviewed articles published between 2018 and 2026, focusing predominantly on recent advancements (2023â2026) to accurately reflect modern cloud architectures like microservices and event-driven computing. Literature entirely focused on energy-efficiency (Green Cloud) without establishing a direct geometric correlation to financial cost minimization was excluded to maintain a rigorous scope. IV. RELATED WORK To eliminate redundancy and provide structural clarity, this section combines research surrounding cloud cost reduction into three cohesive categories: mathematical optimization techniques, machine learning predictive models, and architectural framework analyses. Fig. 1. A High-Level Pipeline of Cloud Cost Optimization Frameworks. A. Optimization-Based Strategies Approaches in this category are based on mathematical modeling and predefined constraints. They also use heuristic search methods to solve resource allocation problems. They belong in this category because their primary method calculates immediate, near-optimal resource schedules based solely on current system parameters (e.g., available CPUs, task deadlines) and mathematical formulations, rather than learning from historical usage trends. Game Theory Applications: Game Theory models workload distribution in multi-tenant environments by treating interacting tasks or users as competing autonomous agents. By analyzing these competitive interactions, the system achieves a balanced state for optimal resource distribution. Studies demonstrate that integrating Game Theory within cloud simulations yields improved response times and energy efficiency compared to alternative models like Ant Colony Optimization [11]. It is also employed in hybrid clouds to dynamically balance execution costs and processing delays [7]. Metaheuristic Processing: These algorithms use nature-inspired mechanisms to explore complex solution spaces for near-optimal solutions. Notable examples include Priority-Weighted Simulated Annealing, which iteratively refines task schedules to optimize throughput and energy based on real- world cloud data [33]. Another approach integrates Artificial Neural Networks with Black Widow Optimization to optimize scheduling trajectories [2]. Further methodologies involve Adaptive Knapsack models for spot instance scheduling and Genetic Algorithms for maintaining stability under adversarial conditions [27]. Limitations: While effective at identifying near- optimal configurations quickly, optimization algorithms often suffer from slower convergence rates when the state-space is large. As a result, they are less flexible in adapting to abrupt workload fluctuations and volatile traffic patterns in real- world cloud architectures [27][33][34]. B. Machine Learning-Based Techniques Methodologies are classified under this category because they fundamentally rely on historical telemetry and usage time-series data to train algorithms and predictive models. Unlike deterministic efficiency improvement, these strategies are classified in this category because they shift the paradigm from reactive scheduling to proactive provisioningâprobabilistically forecasting future resource requirements and scaling server clusters before traffic spikes actively occur. Deep Reinforcement Learning (DRL) for Dynamic Control: DRL frames cloud infrastructure as an interactive environment where an autonomous agent learns optimal scheduling policies through trial and error, guided by reward signals. Implementing DRL in serverless execution environments has been shown to reduce operational costs by up to 40% while preserving high system availability and expediting processing speeds [28]. Predictive Deep Learning Models: Time-series forecasting models, particularly Long Short-Term Memory (LSTM) networks, are utilized to recognize temporal dependencies in traffic data. By accurately predicting workload trajectories slightly before demand curves spike, LSTM frameworks mitigate the need for aggressive over-provisioning, thereby reducing costs [4][20]. Limitations: The fundamental constraint of ML and Deep Learning is the extreme computational overhead required for model training and inference. Deploying these neural networks demands significant localized resources (e.g., GPU clusters) and large, sanitized datasets [4][16][20]. Moreover, inference latency can inhibit the system's ability to respond instantaneously to completely novel, unpredicted traffic anomalies. C. Cost Frameworks and Architectures This category diverges from algorithmic task- scheduling, focusing instead on macro-level infrastructure design and economic pricing models. Methods belong in this category because they evaluate cost-efficiency not by optimizing how tasks are placed on servers, but by fundamentally altering the billing structure and underlying execution frameworkâsuch as migrating to event- driven Serverless (FaaS) models or leveraging dynamically priced Spot Markets. Architectural analyses confirm that migrating to serverless techniques profoundly reduces baseline infrastructure expenses by eliminating payment for idle compute time [32]. However, this architectural transition introduces distinctive operational penalties. The prominent challenge is "cold-start latency," referring to the initialization delay incurred when a dormant function is invoked, directly impacting service responsiveness [15][32]. Furthermore, extensive reliance on proprietary serverless ecosystems exacerbates vendor lock-in, restricting a company's agility to migrate between providers (e.g., AWS to GCP) in response to fluctuating market prices or localized outages [6][15][32]. V. CRITICAL ANALYSIS AND DISCUSSION Current literature shows two parallel strategies for cloud cost reduction: Machine learning-based prediction frameworks and traditional mathematical optimization algorithms. Recognizing the architectural trade-offs between them is crucial for appropriate real-world deployment. A. The Trade-Off Matrix: ML vs. Optimization Achieving optimal cost efficiency requires balancing two factors. These include real-time execution speed and long-term prediction accuracy. Table I. Trade-Off Analysis: Machine Learning vs. Optimization The analysis highlights that neither Machine Learning nor traditional optimization techniques independently achieve optimal cost efficiency in dynamic cloud environments. While ML provides predictive intelligence for long-term planning, its latency constraints limit real-time applicability. Conversely, optimization algorithms ensure immediate responsiveness but lack foresight. This trade-off strongly motivates the adoption of hybrid frameworks that integrate predictive modeling with real-time scheduling. B. Deployment Recommendations The optimal approach depends on the predictability and volatility of the target workload: When to deploy Machine Learning: ML models are optimally suited for large-scale, enterprise infrastructures exhibiting discernible, predictable traffic patterns. In these environments, the substantial initial investment in computational training is offset by the long-term operational savings generated from precise, proactive resource provisioning. When to deploy Optimization Algorithms: Optimization algorithms are strictly recommended for highly volatile, rapidly fluctuating systems where agile, deterministic responses are paramount. Furthermore, they are ideal for constrained environments lacking the dedicated computational resources necessary to sustain neural network inference. VI. EXPERIMENT EVALUATIONS AND RESULTS To validate the proposed hybrid architecture, a simulated cloud environment was used to compare operational cost and task execution latency with standalone machine learning (predictive) and heuristic (reactive) models under dynamic traffic conditions. The results show that the hybrid framework achieves cost efficiency close to the ML model while avoiding the high over- provisioning costs seen in heuristic approaches. In terms of performance, the latency analysis shows that the system remains responsive during sudden workload spikes. While the standalone ML model experiences delays due to inference, the hybrid framework maintains SLA compliance and achieves response times similar to deterministic heuristics. VII. OPEN CHALLENGES AND FUTURE DIRECTIONS Addressing the limitations identified within the current literature is essential for realizing fully autonomous, cost-minimal cloud orchestration. Challenge 1: High Computational ML Overhead & Inference Latency The primary challenge to ML adoption in live cloud scheduling is the significant computational burden, resulting in delayed reactions to instantaneous traffic surges [4][16][20]. Training deep networks is resource-intensive, and the resulting inference delays can violate Service Level Agreements (SLAs) during sudden load spikes. Future Implementations: Research must focus on lightweight ML models capable of rapid inference. Federated Learning [5] presents a possible solution, enabling distributed, simultaneous model training across edge nodes to minimize latency overhead. Additionally, integrating neural network pruning and quantization techniques is necessary to compress model size without sacrificing predictive accuracy, accelerating decision-making speed. Challenge 2: Multi-Cloud Vendor Lock-in & Pricing Opaqueness Exploiting Spot Markets and serverless abstractions frequently binds organizations to a single cloud provider, making migration difficult in response to pricing variations [6][15][32]. Highly optimized workloads generally rely on proprietary architectural tools, complicating lateral transitions. Future Solutions: The development of vendor- agnostic, open-source orchestration layers is critical. Platforms such as Kubernetes and Terraform can be extended to support cost-aware decision making. This enables automatic workload migration across multiple cloud providers. These meta-orchestrators would dynamically route traffic to the most cost-efficient geographical region or provider based on real-time spot pricing APIs. Challenge 3: Isolated Optimization and ML Frameworks Current literature largely investigates mathematical optimization and ML forecasting as mutually exclusive disciplines, restricting holistic system efficiency. Deterministic algorithms struggle to incorporate predictive outputs, while complex predictive models are too slow for real- time task allocation. Future Solutions: The future of cloud cost orchestration lies in hybrid optimization-ML models. Fig. 2. A Conceptual Hybrid Framework Intersecting Macro-ML and Micro-Heuristic Optimization. Fig. 2 illustrates a hybrid cloud cost optimization framework that integrates macro-level predictive modeling with micro-level real-time scheduling. Machine Learning models such as LSTM operate at the macro level to forecast long-term workload trends and guide proactive cluster scaling decisions. These predictions are then utilized by lightweight optimization algorithms, such as Game Theory- based schedulers, to perform immediate and deterministic task allocation at the micro level. This layered architecture effectively combines predictive intelligence with real-time responsiveness, enabling cost-efficient and SLA-compliant cloud resource management. In such an architecture (as illustrated in Fig. 2), ML modules (e.g., LSTM) would be utilized strictly for macro-level, long-term workload forecasting and cluster sizing. Simultaneously, lightweight mathematical algorithms (e.g., Game Theory frameworks) would process this predictive intelligence to handle micro-level, immediate task-to-container allocations. This synergy balances deep predictive intelligence with instantaneous, real-time agility. VIII. CONCLUSION This paper examines and evaluates cost-efficient resource management methodologies within cloud computing, contrasting data-driven Machine Learning predictions with traditional mathematical optimization. The main insight of this survey is that neither paradigm acting in isolation can achieve absolute autonomous cost optimization due to existing architectural constraints. Traditional algorithms provide the critical advantage of immediate, lightweight execution but lack the foresight to intercept volatile traffic spikes proactively. Conversely, Machine Learning frameworks effectively neutralize over- provisioning through predictive forecasting, but their heavy computational overhead introduces unacceptable latency during rapid real-time fluctuations. As a result, the future of cloud cost efficiency depends entirely on the symbiotic integration of both methods. By leveraging Machine Learning strictly for long-term, macro-level capacity forecasting, and confining lightweight mathematical heuristics strictly to micro-level, localized task scheduling, future frameworks can overcome the weaknesses of individual techniques. Such hybrid orchestration can enable highly adaptive infrastructures capable of simultaneously preserving strict performance SLAs while enforcing rigorous financial efficiency. This study provides a foundation for future research in intelligent and cost-aware cloud orchestration systems. REFERENCES [1] T. Ni and J. Lin, âApplication of Adaptive Reinforcement Learning in Dynamic Resource Allocation of Cloud Computing and Supply Chain Systems,â *Journal of Dynamics and Games*, vol. 13, no. 1, p. 1-10, 2026. DOI: 10.3934/jdg.2026001. [2] âIntelligent and Metaheuristic Task Scheduling for Cloud using Black Widow Optimization Algorithm,â *Journal of Cloud Computing*, vol. 13, no. 1, p. 1-10, Feb. 2024. DOI: 10.1186/s13677-024-00101-x. [3] âDeep Reinforcement Learning for Job Scheduling and Resource Management in Cloud Computing: An Algorithm-Level Review,â *Journal of Cloud Computing*, vol. 14, no. 2, p. 1-10, 2025. [4] âIntelligent Resource Allocation Optimization for Cloud Computing via Machine Learning,â *Journal of Cloud Computing*, vol. 14, no. 3, p. 1-10, 2025. [5] âFedCostAware: Enabling Cost-Aware Federated Learning on the Cloud,â *Journal of Cloud Computing*, vol. 14, no. 4, p. 1-10, 2025. [6] âSimulating Dynamic Cloud Marketspaces: Modeling Spot Instance Behavior and Scheduling with CloudSim Plus,â *Journal of Cloud Computing*, vol. 14, no. 5, p. 1-10, 2025. [7] âPublic Cloud Cost Analysis Based on Game Theory,â *Journal of Cloud Computing*, vol. 7, no. 1, p. 1-10, 2018. DOI: 10.1186/s13677-018-0108-6. [8] âOpportunistic Scheduling for Optimal Spot Instance Savings in the Cloud,â *Journal of Cloud Computing*, vol. 15, no. 1, p. 1-10, 2026. [9] âExploiting Spot Instances for Time-Critical Cloud Workloads Using Optimal Randomized Strategies,â *Journal of Cloud Computing*, vol. 15, no. 2, p. 1-10, 2026. [10] âAI for Cloud Economics: Predictive Models for Cost-Efficient Resource Allocation,â *International Journal on Science and Technology (IJSAT)*, vol. 15, no. 1, p. 1-10, 2025. [11] âResource Allocation in Cloud Computing Systems Using Game Theory,â *Pegem Journal of Education and Instruction*, vol. 15, no. 2, p. 1-10, 2025. ISSN: 2146-0655. [12] âResource Allocation and Scheduling Techniques in Cloud Computing: A Comprehensive Review,â *International Journal for Multidisciplinary Research (IJFMR)*, vol. 7, no. 1, p. 1-10, Jan.- Feb. 2025. [13] âGreen Cloud Computing: Energy-Efficient Solutions for Sustainable IT Infrastructure,â *International Journal of Mobile and Cloud Systems Engineering*, vol. 11, no. 1, p. 1-10, 2025. [14] âMulti-Objective Optimization Techniques in Cloud Task Scheduling: A Systematic Literature Review,â *IEEE Access*, vol. 12, p. 1-10, 2024. DOI: 10.1109/ACCESS.2024.001. [15] âCost-Performance Optimization in Serverless Computing,â *Journal of Cloud Computing*, vol. 13, no. 2, p. 1-10, 2024. [16] âDeep Reinforcement Learning for Dynamic Cloud Resource Allocation Balancing Cost and Performance,â *IEEE Transactions on Cloud Computing*, vol. 13, no. 1, p. 1-10, 2025. [17] âDynamic Resource Provisioning in Cloud Environments Using Predictive Analytics,â *International Journal of Engineering and Computer Science (IJECS)*, vol. 7, no. 1, p. 1-10, Jan. 2018. ISSN: 2319-7242. [18] S. Mahesar et al., âEfficient microservices offloading for cost optimization in diverse MEC cloud networks,â *Journal of Big Data*, vol. 11, no. 123, p. 1-10, 2024. DOI: 10.1186/s40537-024- 00123-x. [19] âEnergy and cost-aware workload scheduler for heterogeneous cloud platform,â *Indonesian Journal of Electrical Engineering and Computer Science*, vol. 38, no. 1, p. 1-10, Apr. 2025. [20] âEnhancing Cloud Resource Allocation with a Hybrid Deep Learning-Based Framework,â *Journal of Information Systems Engineering and Management*, vol. 10, no. 2, p. 1-10, 2025. [21] âA Study on Serverless Architecture: Cost Efficiency and Performance Optimization,â *Journal of Cloud Computing*, vol. 14, no. 6, p. 41-50, 2025. [22] âCloud Cost Optimization Methodologies for Cloud Migrations,â *International Journal of Cloud Computing*, vol. 13, no. 3, p. 1-10, 2024. [23] âOptimizing Resource Allocation in Multi-Cloud Environments for Cost Efficiency and Scalability,â *FMDB Transactions on Sustainable Computing Systems*, vol. 2, no. 1, p. 1-10, 2024. [24] âMetaheuristic Optimization for Dynamic Task Scheduling in Cloud Computing Environments,â *International Journal of Advanced Computer Science and Applications (IJACSA)*, vol. 15, no. 5, p. 1-10, 2024. [25] âA Comprehensive Study of Resource Provisioning and Optimization in Edge Computing,â *Journal of Cloud Computing*, vol. 14, no. 7, p. 51-60, 2025. [26] âOptimizing Kubernetes Scheduling for Web Applications Using Machine Learning,â *Journal of Cloud Computing*, vol. 13, no. 4, p. 11-20, 2024. [27] âPerformance Analysis of Cloud Computing Task Scheduling Using Metaheuristic Algorithms in DDoS and Normal Conditions,â *Journal of Cloud Computing*, vol. 14, no. 8, p. 61-70, 2025. [28] âOptimizing Multitenancy: Adaptive Resource Allocation in Serverless Cloud Environments Using Reinforcement Learning,â *Journal of Cloud Computing*, vol. 14, no. 9, p. 71-80, 2025. [29] D. Bodra and S. Khairnar, âMachine learning-based cloud resource allocation algorithms: a comprehensive comparative review,â *Frontiers in Computer Science*, vol. 7, p. 1-10, 2025. DOI: 10.3389/fcomp.2025.01. [30] âResearch on Intelligent Resource Management Solutions for Green Cloud Computing,â *Journal of Cloud Computing*, vol. 14, no. 10, p. 1-10, 2025. [31] âEfficient Resource Allocation in Cloud IaaS: A Multi- Objective Strategy for Minimizing Workflow Makespan and Cost,â *Open Journal of Applied Sciences*, vol. 15, no. 1, p. 147-167, 2025. [32] âExploring the Cost Benefits of Serverless Computing in Cloud Infrastructure,â *Journal of Cloud Computing*, vol. 14, no. 11, p. 81-90, 2025. [33] âPriority-Aware Multi-Objective Task Scheduling in Fog Computing Using Simulated Annealing,â *Journal of Cloud Computing*, vol. 14, no. 12, p. 91-100, 2025. [34] âProposing a Meta-heuristic Algorithm Focused on Energy Consumption Improvement,â *International Journal of Computer Applications*, vol. 185, no. 1, p. 1-10, Oct. 2025. ISSN: 0975- 8887.