Prediction of Resource Utilisation in Cloud Computing Using Machine

Learning

Ruksar Shaikh, Cristina Hava Muntean and Shaguna Gupta

School of Computing, National College of Ireland, Dublin, Ireland

Keywords:

Cloud Computing, Machine Learning, Deep Learning, Resource Utilization, CPU Utilization, Network

Transmission Throughput.

Abstract:

In today’s modern computing infrastructure, cloud computing has emerged as a pivotal paradigm that offers

scalability and ﬂexibility to satisfy the demands of a wide variety of speciﬁc applications. Maintaining optimal

performance and cost-effectiveness inside cloud settings continues to be a signiﬁcant problem and one of the

most important challenges is efﬁcient resource utilisation. A resource utilization prediction system is required

to aid the resource allocator in providing optimal resource allocation. Accurate prediction is difﬁcult in such

a dynamic resource utilisation. The applications of machine learning techniques are the primary emphasis of

this research project which aims to predict resource utilisation in cloud computing systems. The dataset GWA-

T-12 Bitbrains have provided the data of timestamp, cpu usage, network transmitted throughput and Microsoft

Azure traces has provided the data of cpu usage of a cloud server. To predict VM workloads based on CPU

utilization, machine learning models such as Linear Regression, Decision Tree Regression, Gradient Boosting

Regression, and Support Vector Regression are used. In addition to these, deep learning models such as Long

Short-Term Memory and Bi-directional Long Short-Term Memory have also been evaluated in our approach.

Bi-directional Long Short Term Memory approach is considered more effective as compared to other models

in terms of CPU Utilisation and Network Transmitted Throughput as its R2 score is close to 1 and hence can

produce more accurate results.

1 INTRODUCTION

Cloud service providers often adopt a pay-as-you-go

pricing model, which can result in cost savings and

increased ﬂexibility for cloud users. The vast variety

of improvements in cloud computing technology has

resulted in a considerable growth in cloud users and

the development of cloud-based applications to access

various cloud computing services. Several scientiﬁc

applications use cloud computing services, resulting

in varying utilisation of cloud resources. As a result,

efﬁcient resource management is required to handle

the shifting demand of users. Efﬁcient resource man-

agement in a cloud computing environment can help

to optimise resource utilisation, save costs, and im-

prove performance. Resource utilisation prediction

is used to accomplish efﬁcient resource management

(Malik et al., 2022). Predicting the consumption of

cloud resources such as CPU, memory, and network

throughput is critical for effective resource manage-

ment (Kaur et al., 2019). CPU utilisation is one of the

https://orcid.org/0000-0002-9361-3097

most essential metrics for measuring the performance

of host machines. It is also a prominent indicator for

researchers to evaluate when attempting to anticipate

the performance of hosts in the future. The central

processing unit (CPU) is typically the resource that

is subject to the highest amount of demand in virtu-

alized settings. As a result, it is a signiﬁcant con-

tributor to resource shortages on cloud host devices

(Mason et al., 2018). Machine learning algorithms

have gained a lot of attention and are becoming com-

monplace in cloud computing applications in recent

years. Inspired by the structure of the brain, the Neu-

ral Network is one of the most versatile and success-

ful machine learning techniques available. Because

neural networks approximate functions, they can be

used to solve a wide range of issues, from robotics

to regression (Duggan et al., 2017). Efﬁcient re-

source utilisation is still a major difﬁculty in modern

cloud computing settings, affecting cloud systems’

cost-effectiveness and performance. It is difﬁcult to

forecast resource utilisation in such dynamic contexts,

even with cloud models’ inherent scalability and ﬂex-

Shaikh, R., Muntean, C. and Gupta, S.

Prediction of Resource Utilisation in Cloud Computing Using Machine Learning.

DOI: 10.5220/0012742200003711

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

In Proceedings of the 14th International Conference on Cloud Computing and Services Science (CLOSER 2024), pages 103-114

ISBN: 978-989-758-701-6; ISSN: 2184-5042

103

ibility. In this paper, we predict virtual machine CPU

utilization using ML and DL predictive models. This

research aims to investigate the accuracy of predic-

tive models for predicting CPU utilization and net-

work throughput transmission and comparing them.

This paper is discussed as follows: Section 2

presents the overview of the existing works related to

the prediction of resource utilization using ML and Dl

models. Section 3 presents the method of research,

implementation steps of the predictive models. Sec-

tion 4 presents evaluated results and section 5 presents

conclusion and future of the research.

2 LITERATURE REVIEW

Resource usage prediction is becoming more and

more popular due to recent advancements in the ﬁeld

of resource management (Amiri and Mohammad-

Khanli, 2017). The various prediction techniques

based on deep learning and machine learning method-

ologies are compiled in this section.

2.1 Resource Utilisation Using Machine

Learning Techniques

(Borkowski et al., 2016) introduces Cloud resource

provisioning through the use of machine learning-

based models to predict resource utilisation at the task

and resource levels. Evaluations demonstrate signif-

icant gains in accuracy, with 20% reduction in pre-

diction errors and up to 89% improvements. (Con-

forto et al., 2017) presents a unique machine learning-

based resource utilisation prediction system for IaaS

clouds that dynamically estimates resource require-

ments. It offers major improvements in IaaS infras-

tructure management and optimisation by combining

historical data with real-time monitoring to optimise

resource allocation, increase cost efﬁciency, and im-

prove overall IaaS performance.

(Mehmood et al., 2018) emphasises how crucial it

is to allocate resources precisely on cloud platforms to

prevent waste or deterioration in service. It suggests

utilising machine learning approaches to build pre-

cise predictive models for an ensemble-based work-

load prediction system. In large-scale production,

(Morariu et al., 2020) investigates how machine learn-

ing might improve scheduling and resource alloca-

tion. Making use of previous data to develop pre-

diction models, it tackles the intricacies of indus-

trial operations. Learning from past trends, these

models—which include supervised and unsupervised

machine learning algorithms—optimize scheduling

choices. (Daid et al., 2021) investigates data centre

scheduling, with a focus on optimising CPU utilisa-

tion and using machine learning (ML) to fulﬁl service

level agreement (SLA) needs. The paper focuses into

issues with CPU efﬁciency and SLA fulﬁlment and

suggests a hybrid machine learning strategy that com-

bines regression and clustering models for schedul-

ing.

(Manam et al., 2023) suggests a unique method for

cloud computing that optimises resource scheduling

and lowers costs by using the Random Forest algo-

rithm. The approach builds decision trees for classiﬁ-

cation and regression and is well-known for ensemble

learning. A novel approach to predicting CPU uti-

lization in virtualized environments is presented by

(Estrada et al., 2023). It increases forecast accuracy

by clustering related virtual machines according to re-

source utilisation trends using a streamlined VM clus-

tering technique. (Khurana et al., 2023) focuses on

improving Gradient Boosting models to predict CPU

utilisation in cloud environments. This probably en-

tails a lot of parameter optimisation, such as feature

engineering, cross-validation methods, and hyperpa-

rameter ﬁne-tuning.

2.2 Resource Utilisation Using Deep

Learning Techniques

(Wang et al., 2016) introduces a proactive VM de-

ployment approach in cloud computing, using CPU

utilization predictions via the ARIMA-BP neural net-

work. By foreseeing performance issues, it revolu-

tionizes deployment strategies, ensuring service qual-

ity and server efﬁciency. (Duggan et al., 2017) inves-

tigates the use of recurrent neural networks (RNNs)

to predict CPU utilization in cloud computing. By

analyzing historical CPU and network data, the study

employs RNNs to capture temporal dependencies and

forecast usage patterns. (N

as Starberg and Rooth,

2021) focuses on managing CPU ﬂuctuations in cloud

computing by introducing an LSTM model. It fore-

casts CPU usage up to 30 minutes ahead, aiding in dy-

namic capacity scaling. Through performance evalu-

ations against RNNs and state-of-the-art models, its

accuracy in predicting future utilization is assessed.

(Shivakumar et al., 2021) proposes a hybrid model

for cloud resource utilization forecasting, combining

SARIMA for seasonal workloads and LSTM/ARIMA

for non-seasonal patterns. It highlights LSTM’s ac-

curacy in irregular patterns, SARIMA’s effective-

ness in forecasting future usage, and its signiﬁcance

in helping providers avoid resource over or under-

provisioning.

In Table 1 Review of works related to Resource

Utilization Prediction Techniques.

CLOSER 2024 - 14th International Conference on Cloud Computing and Services Science

104

Table 1: Summarized related works of resource utilisation in cloud computing.

Author Title Dataset Tool Technique Result

(Duggan

et al., 2017)

Predicting host CPU

utilization in cloud

computing using re-

current neural net-

works

No applica-

tion/ Dataset of

CoMon project

PlanetLab Recurrent Neu-

ral Network

Prediction

accuracy is

improved.

(Daid et al.,

2021)

An effective schedul-

ing in data centres

for efﬁcient CPU us-

age and service level

agreement fulﬁlment

using machine learn-

ing

Randomly gener-

ated data

Matlab Linear Regres-

sion

Prediction

accuracy is

improved.

(Manam

et al., 2023)

A Machine Learn-

ing Approach to Re-

source Management

in Cloud Computing

Environments

Materna dataset

Trace 3

Google Colabo-

ratory platform

Random Forest

algorithm

Prediction

accuracy is

improved.

(Mehmood

et al., 2018)

Prediction Of Cloud

Computing Resource

Utilization

Google cluster

usage trace data

Cloud system Ensemble

based work-

load prediction

mechanism

Prediction

accuracy is

improved.

(Shivakumar

et al., 2021)

Resource Utilization

Prediction in Cloud

Computing using

Hybrid Model

Bitbrains dataset Experiment was

conducted using

fastStorage, real

trace data of

Bitbrains data

center

SARIMA,

LSTM,

ARIMA

Prediction

accuracy is

improved.

(Conforto

et al., 2017)

Adaptive Resource

Utilization Predic-

tion System for

Infrastructure as a

Service Cloud

Bitbrains dataset fastStorage of

Bitbrains data

center

ARIMA and

Autoregressive

Neural Net-

work (AR-NN)

Prediction

accuracy is

improved.

(Wang et al.,

2016)

Research on the

Prediction Model

of CPU Utilization

Based on ARIMA-

BP Neural Network

IBM Server Xen System ARIMA-BP

neural network

Prediction

can be

improved.

as Star-

berg and

Rooth, 2021)

Predicting a busi-

ness application’s

cloud server CPU

utilization using the

machine learning

model LSTM

Afry dataset Python LSTM Prediction

accuracy is

improved.

Prediction of Resource Utilisation in Cloud Computing Using Machine Learning

105

The Table 1 showcases a variety of approaches

leveraging different datasets, tools, and machine

learning algorithms such as recurrent neural net-

works, linear regression, random forests, and LSTM

among others. Several studies demonstrate improved

prediction accuracy when forecasting resource uti-

lization in cloud environments. However, a com-

pelling trend surfaces from the reviewed literature:

the utilization of LSTM-based models consistently

demonstrates enhanced predictive capabilities across

various datasets. The Bidirectional LSTM, with its

ability to capture long-term dependencies and process

sequential data bidirectionally, presents itself as a ro-

bust choice for modeling the complex temporal pat-

terns inherent in cloud resource usage.

The choice of BiLSTM model stems from its ca-

pacity to effectively capture both past and future con-

text, which is particularly relevant in resource utiliza-

tion forecasting where historical trends and future be-

havior signiﬁcantly impact predictions. The utiliza-

tion of this model offers the potential to enhance ac-

curacy, thereby aiding in proactive resource allocation

and optimization in cloud environments.

3 RESEARCH METHODOLOGY

The research methodology followed in this research

consists of the following steps:

• Research Understanding: With a focus on op-

timising resource utilisation, enhancing perfor-

mance, and cost reduction, the study aims to pre-

dict the accuracy of VM CPU Utilisation and Net-

work Transmission Throughput using ML and DL

prediction models in cloud computing environ-

ments.

• Data Collection: Both qualitative and quantita-

tive information on virtual machine CPU utili-

sation and network transmission throughput was

taken from open-source repositories (BitsBrain

dataset from gwa-t-12-bitbrains, Microsoft Azure

traces from GitHub).

• Data Pre-processing: For model readiness, fea-

ture engineering and data cleaning are performed.

Addressing missing values and getting datasets

prepared for the training of ML and DL predic-

tion models.

• Predictive Models Creation: Using the selected

datasets, train the predictive ML and DL model.

BiLSTM is chosen for its capacity to capture com-

plicated temporal correlations, aiming to predict

VM CPU Utilisation and Network Transmission

Throughput accurately.

• Evaluation: Metrics include Mean Squared Error

(MSE), Mean Absolute Error (MAE), Root Mean

Squared Error, and R-squared (R2) value to assess

prediction accuracy.

• Performance Criteria: Aimed at achieving efﬁ-

cient resource utilization, enhancing performance,

and reducing operational costs.

• Experimentation and Feedback: Experimen-

tal scenarios are meticulously designed with con-

trolled variables to rigorously test the predictive

models.

3.1 Dataset Description

This research utilises two key datasets, the Bitbrains

dataset obtained from gwa-t-12-bitbrains and the Mi-

crosoft Azure Traces 2017 dataset sourced from

GitHub. Both CPU utilisation and network trans-

mission throughput are particularly predicted by the

BitBrains dataset, while the Microsoft Azure Traces

dataset focuses on CPU utilisation in the context of

time series. Timestamp, CPU utilisation, and network

transmission throughput data are the three main pa-

rameters in the BitBrains dataset. Similarly, CPU util-

isation and network transmission throughput patterns

can be predicted using data from the Microsoft Azure

Traces 2017 dataset. Key resource variables, such as

CPU utilisation, are accessible through the datasets

and are essential for the predictive models used in this

research.

3.2 Resource Provisioning Framework

In cloud environments, the availability of comput-

ing resources such as network capabilities, storage

capacities, and CPU power forms the cornerstone

of service provision. Predictive ML and DL mod-

els play a crucial role by forecasting CPU usage

and Network Transmission Throughput, signiﬁcantly

impacting these resources. These models enable

cloud service providers to anticipate resource require-

ments more accurately, thus optimizing the alloca-

tion of network, storage, and CPU resources to align

with projected demand. By harnessing the insights

from these models, cloud environments achieve en-

hanced resource utilization and allocation efﬁciency.

When considering resource allocation strategies, the

inﬂuence of predictive modeling insights is profound

in both reservation-based and on-demand scenarios.

For reservations, predictive models inform alloca-

tion strategies by accurately predicting resource needs

over time. This approach ensures resources are re-

served efﬁciently, minimizing wastage while guaran-

teeing sustained usage in alignment with anticipated

CLOSER 2024 - 14th International Conference on Cloud Computing and Services Science

106

demands. Simultaneously, for on-demand scenarios,

predictive models drive real-time resource optimiza-

tion by dynamically adjusting CPU and network re-

sources based on immediate requirements.

Figure 1: Resource Provisioning Framework.

Optimization strategies, predictive ML, and DL

models play a pivotal role in various aspects of re-

source management. Real-time on-demand resource

optimization leverages these models to swiftly adapt

CPU and network resources to immediate needs. This

agile adjustment signiﬁcantly improves resource uti-

lization efﬁciency, thereby enhancing system perfor-

mance during ﬂuctuating demands. Additionally,

reservation-based optimization harnesses predictive

models to minimize unnecessary resource reserva-

tions, ensuring system stability while maintaining ef-

ﬁciency.

This paper proposes a resource provisioning

model within cloud computing that integrates various

predictive Machine Learning (ML) and Deep Learn-

ing (DL) models, including Linear Regression, De-

cision Tree Regression, Support Vector Regression,

Gradient Boosting Regression, LSTM, and Bi-LSTM.

This model acts as a comprehensive framework, guid-

ing the allocation and management of critical re-

sources like network bandwidth, storage capacities,

and CPU power based on insights derived from these

predictive models in ﬁgure 1.

3.2.1 Linear Regression (LiR)

Linear Regression is a conventional regression model

that seeks to establish a linear correlation between in-

dependent variables (features) and a dependent vari-

able (resource utilization). Linear Regression is a

method used to predict CPU utilization and Network-

transmitted throughput in cloud computing. It aims

to identify direct linear relationships between differ-

ent factors that affect resource usage and the actual

utilization levels.

3.2.2 Decision Tree Regression (DTR)

Decision Tree Regression allows resource utilisation

to be predicted by building a tree-like structure based

on data features. Decision Tree Regression would

generate decision rules based on characteristics like

CPU usage and network transmission throughput in

order to forecast resource utilisation levels in the con-

text of cloud resource prediction.

3.2.3 Gradient Boosting Regression (GBR)

Gradient Boosting Regression creates an ensemble of

decision trees and uses error minimization to improve

predictions iteratively. This model integrates several

techniques to improve forecasts in cloud computing

resource prediction by comprehending intricate in-

teractions between various elements inﬂuencing re-

source usage.

3.2.4 Support Vector Regression (SVR)

Support Vector Regression is a method that deter-

mines the most accurate hyperplane to reﬂect the con-

nection between input data and resource utilisation.

Within the realm of cloud computing, it establishes a

multidimensional threshold to predict CPU utilisation

and Network-transmitted throughput by considering

several aspects.

3.2.5 Long Short Term Memory (LSTM)

LSTM, a form of recurrent neural network (RNN),

excels at capturing dependencies in sequence data.

When predicting resource utilisation in cloud comput-

ing, LSTM would focus on temporal patterns in CPU

usage and network-transmitted throughput, recognis-

ing minor changes and trends that emerge over time.

3.2.6 Bi-Direction Long Short Term Memory

(BiLSTM)

BiLSTM enhances LSTM by performing data pro-

cessing in both the forward and backward directions,

enabling the simultaneous capture of both past and

future context. BiLSTM is used in cloud computing

resource prediction to analyse temporal dependencies

in both directions, allowing for a more comprehen-

sive understanding of CPU utilisation and Network-

transmitted trafﬁc patterns. The comparison between

traditional regression models like Linear Regression,

Decision Tree Regression, Gradient Boosting Regres-

sion, and Support Vector Regression with deep learn-

ing models like LSTM and BiLSTM mirrors the eval-

uation of simpler, rule-based approaches against com-

plex, memory-enhanced models in their ability to

Prediction of Resource Utilisation in Cloud Computing Using Machine Learning

107

predict resource utilization patterns in cloud com-

puting, particularly focusing on CPU utilization and

Network-transmitted throughput.

The proposed approach is grounded in utilizing

the BiLSTM (Bidirectional Long Short-Term Mem-

ory) model to improve the accuracy of predictions

for both VM CPU Utilization and Network Transmis-

sion Throughput. Firstly, datasets from Bitsbrain (for

CPU Utilization and Network Transmission Through-

put) and Microsoft Azure Traces 2017 (for CPU Uti-

lization) are collected and meticulously preprocessed

to ensure completeness and relevance in the con-

text of the study. After that, the predictive models

are implemented and rigorously trained using these

datasets. The primary focus lies in assessing key pre-

dictive metrics such as Mean Absolute Error(MAE),

Mean Squared Error(MSE), Root Mean Squared Er-

ror(RMSE), and R2 score to evaluate the model’s ac-

curacy in predicting VM resource utilization.

3.3 Framework Implementation

This research project employs Python programming

language within Google Colab to predict resource uti-

lization. Figure 2 shows the implementation steps.

Figure 2: Roadmap of Implementation.

This study utilizes two datasets to predict resource

utilization: the Bitbrains dataset, which includes CPU

usage, network transmission throughput, and times-

tamp data. Selected the minimum, maximum, and av-

erage CPU utilization, as well as the timestamp, from

the Microsoft Azure Traces dataset. Both datasets

are implemented individually. The Bitbrains dataset

is utilized for predicting CPU utilization and net-

work transmission throughput, whereas the Microsoft

Azure Traces dataset is speciﬁcally employed for pre-

dicting CPU utilisation.

Prior to initiating the modeling process, metic-

ulous checks are conducted to identify any missing

values and address them appropriately, in order to

guarantee the integrity and comprehensiveness of the

datasets. The data normalization process in both the

Bitbrains and Microsoft Azure Traces datasets in-

volves the use of feature scaling algorithms, specif-

ically MinMaxScaler. The essential preprocessing

stage normalizes the attributes, guaranteeing consis-

tency and optimal efﬁciency throughout the training

of the machine learning model.

The implementation phase commences with a

thorough Exploratory Data Analysis (EDA) focused

on gaining a comprehensive understanding of the

datasets. This comprehensive analysis involves

closely examining important aspects such as the

shape, size, data types, mean values, column names,

counts, standard deviations, and the range between

the minimum and maximum values of the dataset.

These statistical insights offer a comprehensive per-

spective on the datasets, which is crucial for subse-

quent modelling.

The machine learning method begins with iden-

tifying the target column, referred to as ’y’, which

will be predicted by the models. In order to stream-

line the process of training and evaluating the model,

the dataset is split into four distinct subsets: X-train,

X-test, y-train, and y-test. The data is split into two

sets using a 90-10 ratio, with 90% given for training

and 10% for testing. It is easy to re-train the ma-

chine learning and deep learning models if the new

dataset contains the predicted parameters i.e. times-

tamp, CPU usage, and network transmitted through-

put.

This enables the efﬁcient execution of many ma-

chine learning and deep learning techniques such as

Linear Regression, Decision Tree Regression, Gradi-

ent Boosting Regression, Support Vector Regression,

Long Short Term Memory, and Bi-directional Long

Short Term Memory.

These algorithms are executed using a robust li-

brary system, which ensures accurate results. In addi-

tion, this method effectively adapts the algorithms by

using the features provided by the library system.

Evaluation metrics such as Mean Square Error,

Mean Absolute Error, R Square Score, and Root

Mean Square Error are calculated to predict the ac-

curacy of each regression model. It is also noted that

if the number of layers in a model i.e. BiLSTM is in-

creased, the accuracy will increase. Speciﬁcally, it is

seen that when we increased the number of layers by

1 then accuracy showed an improvement of 4% ap-

proximately.

Real-time testing is conducted to validate the

models’ effectiveness in practical scenarios using test

data, ensuring their viability and accuracy in a live

cloud environment. Continuous monitoring and op-

timization of these models remain pivotal, allowing

for adjustments based on evolving cloud infrastruc-

CLOSER 2024 - 14th International Conference on Cloud Computing and Services Science

108

ture dynamics and patterns within the datasets. Ul-

timately, this implementation aims to provide a ro-

bust predictive system facilitating efﬁcient resource

allocation, improved performance, and cost reduction

within cloud computing infrastructures.

4 EVALUATION

In this section, the effectiveness of conventional ma-

chine learning algorithms as described in literature

is assessed against the proposed approach. Employ-

ing the Scikit-Learn library, the experiments are con-

ducted on the Google Colaboratory platform, serving

as the environment for training and testing. Four dis-

tinct machine learning algorithms are evaluated: Lin-

ear Regression, Decision Tree Regression, Gradient

Boosting Regression, Support Vector Regression, and

two deep learning models such as LSTM and BiL-

STM.

4.1 Performance Metrics

4.1.1 Root Mean Squared Error (RMSE)

RMSE is a measure of the differences between pre-

dicted values and observed values. It represents the

square root of the average of the squared differences

between the predicted and actual values. The formula

for RMSE is:

RMSE =

∑

i=1

− ˆy

)

(1)

Here, y

represents the actual value, y

represents

the predicted value, and n is the number of samples.

(Sourced from (Chugh, 2020))

4.1.2 R-squared (R2) Score

R2 score represents the proportion of the variance in

the dependent variable that is predictable from the in-

dependent variables. It is calculated as the ratio of the

explained variation to the total variation. The formula

for R2 score is:

= 1 −

∑

i=1

− ˆy

)

∑

i=1

− ¯y)

(2)

Here, y

represents the actual value, y

represents

the predicted value, and n is the number of samples.

(Sourced from (Chugh, 2020))

4.1.3 Mean Squared Error (MSE)

MSE measures the average of the squares of errors

or deviations. It’s calculated by taking the average of

the squared differences between predicted and actual

values. The formula for MSE is:

MSE =

∑

i=1

− ˆy

)

(3)

Here, y

represents the actual value, y

represents

the predicted value, and n is the number of samples.

(Sourced from (Chugh, 2020))

4.1.4 Mean Absolute Error (MAE)

MAE is the average of the absolute differences be-

tween predicted and actual values. It measures the

average magnitude of errors without considering their

direction. The formula for MAE is:

MAE =

∑

i=1

− ˆy

| (4)

Here, y

represents the actual value, y

represents

the predicted value, and n is the number of samples.

(Sourced from (Chugh, 2020))

4.2 Evaluation of Resource Utilisation

for Machine Learning and Deep

Learning Models

In this research project, we have conducted imple-

mentation using two public available datasets i.e Bit-

Brains and Microsoft Azure Traces 2017. We have

evaluated the CPU Utilisation and Network Trans-

mitted Throughput using BitBrains dataset, along-

with CPU Utilisation evaluated for Microsoft Azure

dataset.

4.2.1 Evaluation of CPU Utilisation

The evaluated results of CPU Utilisation using Bit-

Brains dataset are presented in details in Table 2 and

Figure 4. Also for MicroSoft Azure dataset, the evalu-

ated results of CPU Utilization are presented in details

in Table 3 and Figure 5. The MSE, MAE, RMSE and

R2 metrics of the ML and DL algorithms compared

in this paper are shown in Table 2 and Table 3. The

results shows that for RMSE, Decision Tree Regres-

sion and Gradient Boosting Regression algorithms

had higher error values when compared to BiLSTM

and Linear Regression model which performed bet-

ter than the compared models shown in Table 2 and

Table 3. The evaluated results of the CPU utilization

for the prediction and actual values of the machine

learning models are presented in Figure 4 and Fig-

ure 5. Hence, BiLSTM and Linear Regression per-

formed better than the compared approaches followed

by LSTM.

Prediction of Resource Utilisation in Cloud Computing Using Machine Learning

109

Table 2: For BitBrains dataset - Comparison of machine

learning and deep learning algorithms for CPU Utilization

prediction.

Model MSE MAE RMSE R2 score

LiR 0.0027 0.0215 0.0521 0.7794

DTR 0.0099 0.0402 0.0995 0.1951

GBR 0.0055 0.0294 0.0747 0.5465

SVR 0.0030 0.0389 0.0556 0.7488

LSTM 0.0026 0.0233 0.0515 0.7843

BiLSTM 0.0024 0.0224 0.0490 0.8042

Table 3: For Microsoft Azure dataset - Comparison of ma-

chine learning and deep learning algorithms for CPU Uti-

lization prediction.

Model MSE MAE RMSE R2 score

LiR 0.0002 0.0131 0.0169 0.9833

DTR 0.0023 0.0390 0.0480 0.8661

GBR 0.0010 0.0255 0.0321 0.9399

SVR 0.0015 0.0337 0.0388 0.9127

LSTM 0.0009 0.0239 0.0304 0.9462

BiLSTM 0.0004 0.0169 0.0214 0.9732

Figure 4 and ﬁgure 5 illustrates the predictions

for CPU utilization using a range of machine learning

and deep learning techniques, including Linear Re-

gression(LIR), Gradient Boosting Regression(GBR),

Decision Tree Regression(DTR), Support Vector Re-

gression(SVR), Long Short Term Memory(LSTM)

and Bi-directional Long Short Term Memory (BiL-

STM). These predictive models enable accurate for-

casting of the CPU utilization, providing valuable in-

sights into the resource demands and usage patterns

within the cloud environment. Figure 6 presents the

predictions for the network transmission throughput,

utilizing machine learning and deep learning models,

including LIR, GBR, DTR, SVR, LSTM, and BiL-

STM. These predictions offer valuable insights into

the anticipated network throughput trends and pat-

terns within the cloud environments, aiding in the

proactive management and optimization of network

resources.

4.2.2 Evaluation of Network Transmission

Throughput

The evaluated results for network transmission

throughput are presented in Table 4 and Figure 6.

From the results, it can be seen that BiLSTM has

very close values when compared to the actual value.

BiLSTM and Linear Regression have achieved higher

network transmission throughput prediction accuracy

than the compared models with 0.9 and 0.92 for R2

metrics respectively and lower error rates for RMSE

and MAE.

Table 4: For BitBrains dataset - Comparison of machine

learning and deep learning algorithms for Network Trans-

mission Throughput prediction.

Model MSE MAE RMSE R2 score

LiR 0.0012 0.0114 0.0359 0.9256

DTR 0.0043 0.0473 0.0656 0.752

GBR 0.00183 0.0259 0.0428 0.894

SVR 0.0037 0.0495 0.0610 0.786

LSTM 0.0026 0.0232 0.0513 0.848

BiLSTM 0.00172 0.0203 0.0415 0.901

4.2.3 Performance Comparison of ML and DL

Models for Predicting CPU Utilization

In comparing models for CPU Utilization predictions

using BitBrains and Microsoft Azure datasets, the

BiLSTM and Linear Regression model consistently

stood out as the most accurate in Figure 3. Compared

to LSTM, SVR, GBR, and DTR models, BiLSTM

and Linear Regression consistently demonstrated su-

perior performance across both datasets. BiLSTM

has advantages over Linear Regression as its strength

in capturing complex temporal dependencies allowed

for more precise predictions of CPU Utilization dy-

namics. Also, it works better where data might ex-

hibit non-linear patterns. While other models showed

promise to varying degrees, none matched the robust-

ness of BiLSTM in handling the intricacies within

these datasets. This underlines the pivotal role of

model architecture in effectively predicting CPU Uti-

lization across diverse datasets. The new informa-

tion of this research paper is that the BiLSTM shows

consistently better performance for both the chosen

datasets as compared to other models.

Figure 3: Comparison of Prediction of CPU Utilisation.

CLOSER 2024 - 14th International Conference on Cloud Computing and Services Science

110

(a) Plot of the Linear Regression model predicted vs real

value.

(b) Plot of the Gradient Boosting Regression model pre-

dicted vs real value.

vs real value.

(d) Plot of the Support Vector Regression model pre-

dicted vs real value.

(e) Plot of the Long Short Term Memory model pre-

dicted vs real value.

(f) Plot of the Bi-directional Long Short Term Memory

model predicted vs real value.

Figure 4: Prediction of CPU Utilisation using BitBrains dataset.

Prediction of Resource Utilisation in Cloud Computing Using Machine Learning

111

(a) Plot of the Linear Regression model predicted vs real

value.

(b) Plot of the Gradient Boosting Regression model pre-

dicted vs real value.

vs real value.

(d) Plot of the Support Vector Regression model pre-

dicted vs real value.

(e) Plot of the Long Short Term Memory model pre-

dicted vs real value.

(f) Plot of the Bi-directional Long Short Term Memory

model predicted vs real value.

Figure 5: Prediction of CPU Utilisation using Microsoft Azure dataset.

CLOSER 2024 - 14th International Conference on Cloud Computing and Services Science

112

(a) Plot of the Linear Regression model predicted vs real

value.

(b) Plot of the Gradient Boosting Regression model pre-

dicted vs real value.

vs real value.

(d) Plot of the Support Vector Regression model pre-

dicted vs real value.

(e) Plot of the Long Short Term Memory model pre-

dicted vs real value.

(f) Plot of the Bi-directional Long Short Term Memory

model predicted vs real value.

Figure 6: Prediction of Network Transmission Throughput using BitBrains dataset.

Prediction of Resource Utilisation in Cloud Computing Using Machine Learning

113

5 CONCLUSION

This study conducted a comprehensive exploration

into predicting resource utilization within cloud com-

puting frameworks through a diverse range of ma-

chine learning and deep learning models. Python pro-

gramming within Google Colab was utilized along-

side BitBrains and Microsoft Azure datasets, en-

compassing critical metrics such as CPU usage, net-

work transmission throughput, and timestamps. The

ﬁndings strongly emphasized the efﬁcacy of the

Bi-directional Long Short-Term Memory (BiLSTM)

model, surpassing other machine learning algorithms

in accuracy and performance. The achieved R-square

values and Root Mean Square Error (RMSE) metrics

highlight the BiLSTM model’s exceptional predictive

abilities in anticipating resource utilization, offering

pivotal insights for optimizing cloud computing efﬁ-

ciency.

Based on this research, there are a number of in-

teresting directions for further study. Prediction ac-

curacy might be increased even more by investigat-

ing ensemble learning strategies to integrate different

models. A more thorough grasp of resource usage

patterns may be obtained by extending the dataset’s

reach outside BitBrains and Microsoft Azure. Fur-

ther research into other real-time data aspects may

improve prediction accuracy; nevertheless, improv-

ing the models’ interpretability is still a crucial step

towards gaining more profound understanding.

REFERENCES

Amiri, M. and Mohammad-Khanli, L. (2017). Survey on

prediction models of applications for resources provi-

sioning in cloud. Journal of Network and Computer

Applications, 82:93–113.

Borkowski, M., Schulte, S., and Hochreiner, C. (2016). Pre-

dicting cloud resource utilization. In 2016 IEEE/ACM

9th International Conference on Utility and Cloud

Computing (UCC), pages 37–42.

Chugh, A. (2020). Mae, mse, rmse, coefﬁcient of deter-

mination, adjusted r-squared: Which metric is better.

Medium. http://tiny.cc/137ivz.

Conforto, S., Zia Ullah, Q., Hassan, S., and Khan, G. M.

(2017). Adaptive resource utilization prediction sys-

tem for infrastructure as a service cloud. Computa-

tional Intelligence and Neuroscience, 2017:4873459.

Daid, R., Kumar, Y., Hu, Y.-C., and Chen, W.-L. (2021). An

effective scheduling in data centres for efﬁcient cpu

usage and service level agreement fulﬁlment using

machine learning. Connection Science, 33(4):954–

974.

Duggan, M., Mason, K., Duggan, J., Howley, E., and Bar-

rett, E. (2017). Predicting host cpu utilization in cloud

computing using recurrent neural networks. In 2017

12th International Conference for Internet Technology

and Secured Transactions (ICITST), pages 67–72.

Estrada, R., Valeriano, I., and Aizaga, X. (2023). Cpu us-

age prediction model: A simpliﬁed vm clustering ap-

proach. In Conference on Complex, Intelligent, and

Software Intensive Systems, pages 210–221. Springer.

Kaur, G., Bala, A., and Chana, I. (2019). An intelligent

regressive ensemble approach for predicting resource

usage in cloud computing. Journal of Parallel and

Distributed Computing, 123:1–12.

Khurana, S., Sharma, G., and Sharma, B. (2023). A ﬁne

tune hyper parameter gradient boosting model for cpu

utilization prediction in cloud.

Malik, S., Tahir, M., Sardaraz, M., and Alourani, A. (2022).

A resource utilization prediction model for cloud data

centers using evolutionary algorithms and machine

learning techniques. Applied Sciences, 12(4):2160.

Manam, S., Moessner, K., and Asuquo, P. (2023). A

machine learning approach to resource management

in cloud computing environments. In 2023 IEEE

AFRICON, pages 1–6.

Mason, K., Duggan, M., Barrett, E., Duggan, J., and How-

ley, E. (2018). Predicting host cpu utilization in the

cloud using evolutionary neural networks. Future

Generation Computer Systems, 86:162–173.

Mehmood, T., Latif, S., and Malik, S. (2018). Prediction

of cloud computing resource utilization. In 2018 15th

International Conference on Smart Cities: Improving

Quality of Life Using ICT & IoT (HONET-ICT), pages

38–42.

Morariu, C., Morariu, O., R

aileanu, S., and Borangiu, T.

(2020). Machine learning for predictive scheduling

and resource allocation in large scale manufacturing

systems. Computers in Industry, 120:103244.

as Starberg, F. and Rooth, A. (2021). Predicting a busi-

ness application’s cloud server cpu utilization using

the machine learning model lstm.

Shivakumar, B. R., Anupama, K. C., and Ramaiah, N.

(2021). Resource utilization prediction in cloud

computing using hybrid model. International Jour-

nal of Advanced Computer Science and Applications,

12:2021.

Wang, J., Yan, Y., and Guo, J. (2016). Research on the

prediction model of cpu utilization based on arima-

bp neural network. In MATEC Web of Conferences,

volume 65, page 03009. EDP Sciences.

CLOSER 2024 - 14th International Conference on Cloud Computing and Services Science

114