Dynamic Task Scheduling Using Machine Learning and Enhanced

Fuzzy Logic System for Efficient Resource Utilization in Virtual

Cloud

Panchagnula Kamakshi Thai and Shanker Chandre

Department of Computer Science & Artificial Intelligence, School of Computer Science & Artificial Intelligence, SR

University, Warangal, Telangana, India

Keywords: Fuzzy Logic Systems, Task Scheduling, Machine Learning, Virtualized Cloud Systems.

Abstract: Virtual clouds need intelligent task scheduling systems because their limited resources become more efficient

through workload-based scheduling strategies. Fuzzy logic systems offer the best solutions for handling tasks

in cloud computing because they can deal with uncertain situations and changing workloads and resources.

The integration of heuristic interpolated models and machine learning algorithms achieves optimized task

scheduling while distributing resources evenly and shortening execution duration. Machine learning uses

supervised learning to predict resources and reinforcement learning to adjust decisions, helping to construct

flexible and accurate execution patterns. An improved version of fuzzy logic contains smart scheduling

functionality that adapts priority settings based on both mission-dependent needs along external operational

factors such as execution period and urgency level, as well as system resources and system utilization.

Enhanced fuzzy logic systems (EFLS) is one of the models used in the research to automatically change

schedules based on environmental factors and changes in job demand. The system constructs exhaustive

membership functions that show overlapping job priority areas and limits on resources using its method. The

system contains four major modules consisting of submission tracking, resource monitoring alongside

predictive capabilities, and optimized decision management that permits real-time capability. The

performance assessments reveal significant positive outcomes in all three areas: makespan, task completion

rates, and resource utilization as compared to conventional methods. The method demonstrates how

virtualized cloud systems can implement scalable, efficient, adaptive task management.

1 INTRODUCTION

Dynamic task scheduling in a virtual cloud

environment, where the cluster is virtualized,

scalability is achieved, and multi-tenancy and

variability of workload have prevailed, is an

important way to utilize resources. The sociology and

cloud behavior dynamics suggest potential for

improved scheduling frameworks to adapt to

economic conditions and challenges in a competitive

cloud market. The standard scheduling techniques

that operate based on fixed rules show deficiencies

when resource types and workloads require

adjustments. Virtual-based cloud environments

require task scheduling operations to distribute

computational tasks across VMs to achieve optimal

resource utilization as well as energy efficiency while

maximizing makespan and task completion rate.

Virtual-based cloud environments face two main

challenges: varied workloads, differing resource

types, and task dependencies between resources. The

accuracy of decision-making improves when neural

networks and Random forests from machine learning

methods gain increasing utilization for resource need

predictions as well as execution duration predictions

(Zhang et al., 2021). Operational efficiency and

resource productivity grow through implementing

flexible decisions about job scheduling strategies.

Artificially intelligent fuzzy logic systems, together

with machine learning techniques, serve this purpose.

Implementations that unite improved fuzzy logic with

ML surpass traditional application methods because

they deliver multiple advantages. The analysis of

historical data by ML models helps to predict job

execution times, which leads to proactive resource

allocation (Gupta et al., 2023). Classification

algorithms priorities jobs, while regression-based

816

Thai, P. K. and Chandre, S.

Dynamic Task Scheduling Using Machine Learning and Enhanced Fuzzy Logic System for Efﬁcient Resource Utilization in Virtual Cloud.

DOI: 10.5220/0013890400004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 2, pages

816-822

ISBN: 978-989-758-777-1

models forecast resource usage. For precise schedule

optimization, advanced fuzzy logic systems use

environmental data, deadline limits, and real-time

resource availability. The hybrid method balances

efficiency and justice by assigning critical tasks to

resources without overburdening them. A major

benefit of the hybrid architecture is its energy

efficiency. The cloud data center's high energy usage

has financial and environmental consequences. The

system applies ML predictions together with fuzzy

logic rules to decrease resource disuse and select the

right resources for each task (Chen et al., 2022).

The reduction of energy consumption occurs

because energy-aware scheduling algorithms use

real-time energy profiles from resources as part of

their decision-making process. The suggested

framework adopts dynamic approaches to operational

changes, whereas static systems use established

criteria exclusively to operate. Using fuzzy logic

systems along with machine learning models allows

for flexible changes in priority levels and can predict

delays during busy times, helping with replacing

resources. The adaptable nature of this system

ensures high completion rates when task uncertainties

exist, and this creates substantial stability

improvements (Rahman et al., 2021). The

implementation of scheduling systems using

combinations of ML and fuzzy logic generates

important advantages, though it comes with certain

implementation challenges. Table processing speed

rises due to both sophisticated ML model design and

massive training data needs. The development of

fuzzy logic rules needs knowledge about the domain

together with continuous modifications to capture

actual situations accurately. New developments in

automatic fuzzy rule generation together with ML

model optimization have effectively reduced this

challenge (Li, Y., and Wang, T, 2023). The paper

helps exploration scheduling by looking at smart

scheduling methods that use fuzzy logic along with

machine learning. The recent development

demonstrates the outstanding capacity for dynamic

cloud system employment because test results

indicate it boosts resource utilization while

decreasing energy usage while upholding task

execution timing guarantees.

2 RELATED WORKS

In Saad et al., (2023). the authors translated K-Means

clustering through fuzzy logic for the effective

organization of fog nodes by their resource

characteristics and workload patterns. The arrived-at

method distributes work in real time by linking K-

means clustering to fuzzy logic and fuzzy logic

adaptability. Their approach demonstrated how

distributed job placement to fog nodes using machine

learning generated decreased execution times and

reduced response times and network utilization rates.

Thus, extensive testing confirms that the proposed

solution results successful in being versatile in

changing fog scenarios. The time-consuming VM

work cluster detection, but we the entire process is

very efficient. They developed and evaluated their

proposed approach using iFogSim. It shows

distinguished improvements in comparison to

machine learning and non-ML-based scheduling

methods inside the iFogSim framework in terms of

response time, execution time, and lesser network

utilization in the simulation results. In Thapliyal et al.,

(2024), authors proposed an optimized approach

based on fuzzy logic (FL) and best-fit-decreasing

(BFD) for job scheduling process in a cloud

computing environment. They all play into making

FL-BFD worthy of your time, money, power, and

resources. The FL-BFD reallocates the cloud VMs

by the user demand. We find it important to leverage

the FL capabilities to deal with uncertainty and

missing information to properly provision the

needed with what the user requires in the BFD for

properly provisioning VMs. The proposed FL-BFD

inspects multiple factors including makespan,

computational time, degree of imbalance, power

consumption, and SLA violations. Output: FL-BFD

has the longest makespan of 9.2 ms among 1000

jobs, compared to IWHOLF-TSC and MCT-PSO.

The authors of Radhika, D (2022) presented a

cloud dynamic task scheduling in which they consider

big data analysis processing in the cloud

environment. They employed multiple methods,

including a machine learning classifier and an

optimization approach. For classification of various

virtual machine tasks, they use a machine-learning

classifier known as a Support Vector Machine

(SVM). Using this classifier, we can effectively

reduce makespan and execution time when

classifying incoming requests. They also assigned the

classified job using moth flame optimization to the

SVM classifier. This proposed system is used to:

classify the virtual machine (VMs) tasks and

evaluate decision make methodology for the

resources allocation. The proposed method showed

that the make-span time may be reduced, while load

balancing may also seem beneficial according to their

work, which they tested in a cloud modeling

environment to improve VM classification. Iin Alam

et al., (2021), the authors introduce a new static

Dynamic Task Scheduling Using Machine Learning and Enhanced Fuzzy Logic System for Efﬁcient Resource Utilization in Virtual Cloud

817

homogeneity task assignment (ESTA) approach

aimed at optimizing average utilization. Handling of

jobs should therefore be done with efficient allocation

of tasks between the available resources and to

prevent scheduling algorithm challenges. ESTA

algorithm uses the shortest completion time method.

ESTA intelligently maps batches of independent

tasks (cloudlets) on the top of heterogeneous virtual

machines in the IaaS cloud computing to optimize the

use of virtual machines. The performance of ESTA is

analysed against Min-Min,LBSM,LJFR-SJFR,

Sufferage, MCT, MET, and OLB using simulation

study in terms of make span, utilization, and response

time to identify the strengths and weaknesses of

ESTA. In Gong et al., (2024), authors also discussed

importance of deep reinforcement learning and

machine learning in optimizing virtual machine

migrations and managing resources of cloud. Deep

reinforcements learning and other machine learning

techniques are effective for optimizing embodied

intelligence, dynamic resource management, and

environment recognition. All of this is possible

because of their adaptability, policy creation and

forecasting capabilities. Through cloud computing,

organizations using cloud services can optimize

resource utilizations, reduce power consumption, and

improve quality of service delivered, enabling them

to benefit from these technologies.

3 MATERIALS AND METHODS

The enhanced fuzzy logic functions with machine

learning (ML) technique, enable maximum output of

resources usage within virtual cloud environments

using the dynamic task scheduling system proposed

in Figure 1. a Problem solving before execution:

execution time prediction, resource need prediction

and wait time prediction in workflow through

applying machine learning models within a system.

Work urgency, system load, and resources

availability are factors that determine the fuzzy logic

adaptation in scheduling priority. Associated task-to-

resource assignments are obtained through the use of

flexible rules and predictive analytics that address

resource consumption with respect to its constraints

while meeting job deadlines and feasibility

requirements. The proposed solution comprises a data

preprocessing module, ML prediction engine, fuzzy

logic scheduler, and a performance monitoring unit.

When applied to a traditional scheduling approach, it

dramatically improves energy efficiency and job

completion ratios. With this intelligent, scalable

solution, we enable consistent, safe, and efficient

management of complex application workloads

across agile cloud environments built on OpenShift

data to provide a magnificent experience.

3.1 Data Preprocessing Module

It cleans and normalizes the raw input data while

encoding it making it ready for analysis. Data points

for resource measurement status (CPU, memory, etc)

and relevant task properties, e.g. priorities and

deadline, are controlled by the system. Dynamic task

scheduling in virtual cloud environments relies on

effective allocation of tasks to resources. This process

is used to ensure the system is making the best use of

its resources, prioritizing tasks, and responding well

to requests, among other things.

Figure 1: Proposed Dynamic Task Scheduling Framework.

Enhanced fuzzy logic systems improve

scheduling by analyzing work requirements and

adjusting task priorities on the fly. Simultaneously,

ML models predict the duration of each task and the

required resources.

𝑥



=

1, if task 𝑇



is assigned to resource 𝑅



0, otherwise

(1)

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

818

Activities and resources placed in a dynamic

virtual cloud environment are described by certain

criteria to support optimal scheduling. A task (Ti) is

defined by a set of attributes: its central processing

unit (cpui), memory (memi), priority (pi) and

deadline (di). Task execution time and resource

allocation can also be dictated by the size,

complexity, and interdependence of the tasks. Also

important are the maximum consumed CPU (Cj),

memory (Mj), and energy consumption rate (Ej) for

each resource (Rj). You: Being fast at a low price is

nice, but the number of available storage, availability,

and bandwidth also affect how the system will

perform under a variety of loads. Task-to-resource

mapping achieves its objectives based on an

analytical approach taking into consideration the

reduction of wait time and optimization of efficiency

while maintaining the energy constraints. Real-time

workload needs require flexible solutions to be

implemented with advanced fuzzy logic and artificial

intelligence techniques to be deployed effectively

without compromising scalability. Activities, and

resources in a dynamic virtual cloud surrounding are

described with the help of certain features for

facilitating optimal scheduling. Each task (Ti) has

associated factors such as central processing unit

(cpui), memory (memi), priority (pi), and deadline

(di). The execution time {@link Measured Execution

Type #execution Time()} and resource allocation

{@link Measured Execution Type #resource

Allocation()} of tasks can also depend upon their size,

complexity, and interdependence. For each resource

(Rj), the maximum CPU (), memory (), and energy

consumption rate (Ej) are equally important. On the

other hand, more storage, availability, and network

bandwidth impact the system's performance under

varying workloads. Both task-to-resource mapping

achieves its objectives through analytical

consideration of waiting time reduction and

efficiency optimization in addition to energy

constraint management. Real-time workload

demands adaptive solutions that require enhanced

fuzzy logic and machine learning techniques for

scalability and efficiency.

𝑀𝑎𝑥:𝑅 =

∑∑





(









)









∑





















(2)

𝑀𝑖𝑛:𝑀 = max





∑

𝑥



.𝑇



𝑇



,𝑅









 (3)

𝑀𝑖𝑛:𝐸 =

∑∑

𝑥











𝐸



(4)

𝑀𝑎𝑥:𝑃 =

∑∑

𝑥











𝑝



(5)

Virtual cloud environments need multiple

fundamental requirements which ensure resource

efficiency and on-time task completion and maintain

system reliability. The four types of resource

constraints represent some of the key restrictions

within dynamic virtual cloud scheduling systems.

3.2 Resource Constraints

Virtual machines and other resources have limited

CPU, memory, and storage space.

∑

𝑥







.𝑐𝑝𝑢



≤𝑐𝑝𝑢





∑

𝑥







.𝑚𝑒𝑚



≤𝑚𝑒𝑚





(6)

where x

is a binary variable indicating task-

resource assignment.

Task Deadline Constraint: Timely completion of

tasks is essential (d

). With a resource (R

) and a task

), For each task, we allocate a single resource:

𝑇



𝑇



,𝑅



+𝑇



𝑇



,𝑅



≤𝑑



(7)

Task Assignment Constraint: Each task is

assigned to one and only one resource:

∑

𝑥



=1 ,∀𝑖





(8)

3.3 Feasibility Constraint

To allocate resources, they must meet the conditions

for task execution. If resource R

is unable to meet T

needs, then x

equals zero. To predict execution

timing and resource requirements, machine learning

models accept these described attributes.

3.4 ML-Based Prediction Engine

The following task and resource attributes are fed into

ML models to predict when a task will need to be

executed and how many resources will be needed:

𝑇



𝑇



,𝑅



=𝑓𝑐𝑝𝑢



,𝑚𝑒𝑚



,𝑅



) (9)

where 𝑓 is an ML model (e.g., Random Forest, Neural

Network) trained on historical data.

𝑈



𝑅



=𝑔

(

task attributes,current resource state

)

(10)

where g is an ML regression model predicting

resource usage trends.

Dynamic Task Scheduling Using Machine Learning and Enhanced Fuzzy Logic System for Efﬁcient Resource Utilization in Virtual Cloud

819

3.5 Enhanced Fuzzy Logic System

Scheduler

The scheduling system adjusts task priorities while

allocating resources through its enhanced fuzzy logic

model that considers urgency levels and hardware

capacity limits. Real-time job requirements, along

with environmental factors, drive the scheduling

system to both distribute resources effectively and

determine task execution time simultaneously.

Define membership functions for:

𝜇



(

𝑝



)



Low,Medium,High,Critical

(11)

𝜇



 



Capacity

(12)

𝜇



 



Workload

(13)

based on predefined priority levels.

4 RESULTS AND DISCUSSION

To implement the proposed dynamic task scheduling

framework, a robust set of tools and programming

languages is required. This framework combines

advanced fuzzy logic systems with machine learning

(ML). The main reason Python is used so often is

because of its large library support for machine

learning (e.g., Scikit-learn, TensorFlow, PyTorch)

and data analysis (e.g., NumPy, Pandas). Matlab's

flexibility as a modeling and simulation tool

facilitates the simple design and optimization of fuzzy

logic systems. CloudSim allows users to duplicate

procedures of resource allocation distribution

alongside virtual cloud scheduling functionality.

Visualization tools like Matplotlib and Seaborn

produce comprehensive graphs and plots. The

proposed framework receives analysis and quick

processing through Jupyter Notebook as an Integrated

Development Environment (IDE), which simplifies

tests while also improving debugging and the

combination of framework components.

The real-time operational system achieves

decision times that are 25% faster through this model

implementation. Allocation decisions execute well

combined with real-time prioritization because of

resource management and operational adjustment.

The ML-based fuzzy logic model demonstrates

superior performance compared to traditional and

purely ML scheduling approaches, as illustrated in

Table 1.

Table 1: Analysis of Make Span.

Tasks

Conventional

Scheduling

(min)

ML-Based

Scheduling

(min)

ML +

Fuzzy

Logic

(min)

500 120 100 90

1000 240 210 190

1500 358 384 360

2000 480 435 405

2500 495 450 420

Figure 2 make span decrease shows how much

time it takes to finish all tasks when using traditional,

ML-based, and ML with improved fuzzy logic

scheduling methods. It emphasizes that the suggested

hybrid framework significantly reduces makespan as

compared to conventional methods. The hybrid

solution achieves continuous better performance with

increasing tasks because it dynamically adjusts work

priorities alongside resource allocation. Visual

representation by the hybrid technique demonstrates

its ability to extend project durations while reducing

delays to achieve successful completion of tasks in

virtual cloud environments.

Figure 2: Performance Comparison of Makespan Across

Scheduling Methods for Varying Task Counts.

The hybrid strategy optimizes resource utilization

through its dynamic action to move underused

resources and eliminate bottleneck points (Table 2),

which keeps efficiency steady regardless of workload

changes.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

820

Table 2: Analysis of Resource Utilization.

Tasks

Conventional

Scheduling (%)

ML-Based

Scheduling

(%)

ML +

Fuzzy

Logic

(%)

500 70 80 85

1000 65 78 84

1500 60 75 82

2000 56 70 74

2500 48 63 70

A comparison of resource utilization emerges in

Figure 3 regarding the three scheduling approaches,

including traditional, ML-based, and ML-with-

enhanced fuzzy logic scheduling. Task optimization

in combination with dynamic resource allocation,

improves the hybrid strategy because it delivers

sustained high resource utilization values. When task

numbers increase, the hybrid system preserves its

already effective performance level. Such resource

allocation optimizes performance because it

simultaneously minimizes bottlenecks while

maintaining low idle resource conditions. Through

visualization, the framework demonstrates its ability

to distribute cloud resources effectively while

retaining balanced usage under any workload

conditions.

Figure 3: Performance Comparison of Resource Utilization

Across Scheduling Methods.

The hybrid architecture achieves better task

success rates when properly setting high-priority jobs

because of its organizational structure (Table 3).

Table 3: Analysis of Task Completion Rate.

Tasks

Conventional

Scheduling (%)

ML-Based

Scheduling

(%)

ML +

Fuzzy

Logic (%)

500 90 95 98

1000 85 93 96

1500 80 88 92

2000 76 80 94

2500 65 72 90

Figure 4 shows that the standard method, along

with ML-based scheduling and ML-enhanced fuzzy

logic scheduling methods, produced different time-

frame completion metrics. The hybrid methodology

proves capable of finishing jobs speedily throughout

all workload levels. The smart resource distribution

as well as the critical activity prioritization system

enables this improvement. The framework

demonstrates its capability to adapt to changing

workloads thanks to its resource constraints as

reflected by the provided data. The hybrid strategy

succeeds in safeguarding performance dependability

and optimizing work throughput for virtual clouds

because it reduces deadline violations.

Figure 4: Performance Comparison of Task Completion

Rates Across Different Scheduling Approaches.

5 CONCLUSIONS

We suggest the way of resource get optimum

utilization through advance techniques of machine

learning from fuzzy logic systems to virtual cloud

Dynamic Task Scheduling Using Machine Learning and Enhanced Fuzzy Logic System for Efﬁcient Resource Utilization in Virtual Cloud

821

environment, which allows resource optimally usage.

ML's predictive abilities allow this framework to

analyze historical and real-time data to proactively

allocate resources and prioritize jobs. Improved

fuzzy logic systems keep things flexible by

continuously making scheduling decisions that are

optimal based upon work requirements and external

input like system load and resource availability. It

outperforms the previous knowledge and work in

this area, evidenced through crucial performance

metrics with up to 25% makespan savings, 21.43%

resource utilization improvement and 8.89% higher

task completion rates results. The hybrid approach

delivers better scheduling performance than

conventional and standalone ML-based scheduling

approaches by considering dynamic workloads and

complicated resource constraints. The approach

supports both scalability and energy efficiency in

modern cloud systems to ensure optimal performance

in diverse operational environments. Overall, there

are two aspects of the framework which could further

improve it: A/ {a|the dynamic development} on the

basis of reinforcement learning and B/{a|the platform

for the real-world distributed real-time deployments

studies}. This architecture represents a ground-

breaking method for complex job scheduling in

advanced cloud computing clouds.

REFERENCES

Alam, Mahfooz &., Mahak & Haidri, Raza & Yadav, Dr-

Dileep. (2021). Efficient Task Scheduling on Virtual

Machine in Cloud Computing Environment.

International Journal of Pervasive Computing and

Communications. 17. 271- 287. https://doi.org/10.110

8/IJPCC-04-2020-0029.

Chen, J., Zhang, Y., & Liu, W. (2022). Energy-efficient

task scheduling in cloud data centers: A review. Journal

of Cloud Computing, 10(3), 45- 60. https://doi.org/10.

1007/s10115-022-01234-7

Gong, Yulu & Huang, Jiaxin & Liu, Bo & Xu, Jingyu &

Wu, Binbin & Zhang, Yifan. (2024). Dynamic resource

allocation for virtual machine migration optimization

using machine learning. Applied and Computational

Engineering. 57. 1-8. https://doi.org/10.54254/2755-

2721/57/20241348.

Gupta, P., Kumar, R., & Yadav, S. (2023). Machine

learning for predictive resource allocation in cloud

computing. ACM Computing Surveys, 55(4), 1-34.

https://doi.org/10.1145/3545678

Kumar, A., & Singh, H. (2020). Adaptive fuzzy logic for

dynamic scheduling in cloud computing. International

Journal of Computational Intelligence Systems, 13(2),

1180-1194. https://doi.org/10.1007/s00521-019-

04210-8

Li, Y., & Wang, T. (2023). Advances in hybrid scheduling

algorithms for cloud environments. IEEE Transactions

on Cloud Computing, 11(2), 150-165.

https://doi.org/10.1109/TCC.2022.3184927

Radhika, D. (2022). Virtual Machine Task Classification

Using Support Vector Machine and Improved MFO

Based Task Scheduling. Journal of University of

Shanghai for Science and Technology. 24. 1-15.

https://doi.org/10.51201/JUSST/22/0159.

Rahman, M., Alam, M., & Hasan, M. (2021). Dynamic

scheduling in cloud environments using machine

learning: A review. Future Generation Computer

Systems, 125, 567- 584. https://doi.org/10.1016/j.futu

re.2021.03.015

Saad, Muhammad & Qureshi, Rehan. (2023). Machine

learning-driven task scheduling with dynamic K-means

based clustering algorithm using fuzzy logic in FOG

environment.

Thapliyal, Nitin & Dimri, Priti. (2024). Task scheduling

using fuzzy logic with best-fit-decreasing for cloud

computing environment. Cluster Computing. 27. 1-16.

https://doi.org/10.1007/s10586-024-04378-7.

Zhang, X., Chen, Q., & Lin, M. (2021). Task scheduling

with ML-based resource prediction in cloud

environments. Concurrency and Computation: Practice

and Experience, 33(12), e6028. https://doi.org/10.1002

/cpe.6028

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

822