Analyzing Facility Servers Using Random Forest and XGBoost for

Optimized Job Allocation

Potula Radha Nishant

and Beena B M

Department of Computer Science and Engineering, Amrita School of Engineering,

Bangalore, Amrita Vishwa Vidyapeetham, India

Keywords: Carbon Emission, XG Boost, Green Computing, Cross-Value, Energy Consumption, Energy Conservation.

Abstract: Computer systems consume huge amount of energy causing higher levels of carbon emissions thus polluting

the environment. This study addresses the issue by developing machine learning algorithms to conserve

resources across datacentres. The machine learning models have been developed to predict a higher level

accuracy focusing job level scheduling. The Random Forest used for job scheduling may result in enhancing

performance of green data centres by reducing the energy consumption. Our future research tries to improve

the existing resource management solutions focusing on job level characteristics.

1 INTRODUCTION

It is mind-boggling in the way data centers are

regularly linked to contemporary computing, they

massively contribute to carbon footprint due to the

enormous energy used in supporting servers, storage,

as well as networking systems (Selin, 2024). This

high electricity demand is generally produced from

fossil fuels hence partnering large amounts of carbon

dioxide ( CO2). Also, the devices that help to keep

equipment at an appropriate temperature to operate,

worsens energy utilization, thereby making the power

of the facility high.

Diesel generators that are used during black out

also release CO2 further stressing the importance of

clean up practices. The misuse of resources is viewed

to worsen the environmental effects hence inefficient

utilization of resources, energy and other resources

get wasted.

Flexible resource management will improve the

use of resources, minimise any likelihood of resource

wastage, and closely monitor energy consumption.

Others are scope 3 emissions which include emissions

from the following infrastructure, the equipment of

data center and other related equipment. This research

can be related to several Sustain-

https://orcid.org/0009-0009-8855-1643

https://orcid.org/0000-0001-9108-7073

able Development Goals, namely, SDG 3 on clean

air quality, SDG 4 through integrating sustainability

into education curricula and materials, SDG 7 for

affordable and clean energy through green jobs and

innovations in the Cloud technology, as well as SDG

8 and 9 through creation of green jobs and

technological innovations respectively. It also helps

in achieving of sustainable development goal 11

(Sustainable cities and communities), SDG12

(Responsible consumption and production),SDG13

(Climate action), SDG15( Life on land) through the

efficiency of resources and minimizing emissions. In

addition, SDG-17 (Partnerships for the Goals)

implementation is backed by cooperation with

industry and academia. It advances the work by

utilizing XGBoost and Random Forest regression

algorithms for efficient job allocation depending on

rack sensor information. it aims at reducing energy

consumption; reducing carbon footprint; and

enhancing resource allocation efficiency in data

center using machine learning methods. This strategy

looks into several concerns in the environment and

opens doors to improve the sustainability processes

for cloud computing platforms.

Nishant, P. R. and B M, B.

Analyzing Facility Servers Using Random Forest and XGBoost for Optimized Job Allocation.

DOI: 10.5220/0013577100004639

In Proceedings of the 2nd International Conference on Intelligent and Sustainable Power and Energy Systems (ISPES 2024), pages 74-81

ISBN: 978-989-758-756-6

2 LITERATURE SURVEY

This survey focuses upon some of the current

developments in green computing and many energy-

efficient technologies in different areas with

emphasis on innovative techniques for resource

optimization and reduced environmental impact. The

paper gives a thorough survey of fourteen-

cryptography-relevant studies dealing with the

opportunities and challenges in the fields of data

mining, cloud services, and high-performance

computing. They point out that adaptable algorithms

and frameworks are essential in meeting the varying

demands of heterogeneous computer environments,

which in themselves call for sustainability.

2.1 Related Work

Guo et al. (Guo et al., 2023) undertook initial research

on HPC, which exploited sensor data from large-scale

networks to analyze the workload distribution on

energy efficiency, using two techniques- workload

optimization and dynamic core allocation- to

minimize energy and enhance system utilization.

However, these methodologies have problems

regarding multibody systems with diverse

temperature and energy management requirements.

Abbas et al. (Abbas et al., 2023), like Guo et al.,

propose an energy-efficient architecture that depends

on renewable energy sources and consequently one

that favors green computing. Their approach intends

to optimize resource consumption and encourage

sustainable energy use in computing environments.

However, it was deemed, indeed, that creating robust

algorithms that can adapt dynamically to diverse

energy sources is essential for accomplishing

sustainability as well as optimal performance.

Ahmad et al. (Ahmad et al., 2021) carried out an

encompassing literature review in order to find out

the practices and challenges brought by adopting

green cloud computing, but from a client-centric

point. According to their findings, sustainable

practices have to be incorporated in cloud computing

to help lessen the impacts of energy consumption,

environmental responsibility, and reliability of the

services. The creation of complete frameworks

considering the sustainability of hybrid cloud

services, including qualitative studies to consider

their environmental influence, together with

validation of proposed green techniques, remains

open.

Within the field of mobile cloud computing,

Skourletopoulos et al. (Skourletopoulos et al., 2018)

introduce a model of elasticity debt analytics that

aims to optimize resource provisioning, employing a

game- theoretic approach to reduce elasticity debt.

These techniques remain a real challenge in adapting

the model to changing conditions and integrating ML

technologies for enhanced resource utilization.

Raja (Raja, 2021) explains how green computing

can reduce energy waste in the IT sector and further

other approaches to minimize carbon footprints, such

as through energy-efficient data centers and

renewable energy sources. He discusses the potential

of greening initiatives with respect to environmental

sustainability for the IT sector, while flexible

management and control over varied energy demands

will specifically require adaptive solutions.

Qiu et al. (Qiu et al., 2018) discuses on

exploration of how Cloud Service Brokers might

provide new avenues toward energy efficiency and

quality of service through optimized demand

allocation and pricing strategies. While the work by

these authors shows some improvement over that by

others, they still face challenges with real-world

deployment and scalability issues.

Qiu et al. (Qiu et al., 2015) also give an insight

into PCM optimization in Green Cloud Computing

using genetic algorithms aimed at improving memory

usage and efficient resource allocation.

Tuli et al. (Beloglazov and Buyya, 2014)

proposed an energy-aware combinatorial virtual

machine allocation model for minimizing the power

consumption in data centers. This model works well

in static circumstances but the architecture is hemmed

in by open issues regarding the management of

workloads for real-time contexts and requires

adaptive algorithms to scale up with emerging

technologies such as edge computing and IoT.

Alarifi et al. (Xiao and Li, 2018) suggest an

Energy-Effective Hybrid framework for cloud data

centers that differently consolidate and utilize servers.

However, optimization of migration algorithms and

transition to sustainable energy sources are some

open issues still facing researchers in this area.

Chiaraviglio et al. (Chiaraviglio et al., 2014) put

forth a dynamic methodology for online power and

load computation, whereby the server’s power states

can be dynamically altered. This will result in a very

high saving in energy needs. However, many open

problems relating to scalability and multi-objective

optimization remain open.

Kulkarni et al. (Kulkarni et al., 2024) continue

with innovation and creation of cloud-based mood-

driven music recommendation system combining

personalized recommendations from user profiles,

collaborative filtering, and machine learning. The

system, with its scalable architecture is an apt

Analyzing Facility Servers Using Random Forest and XGBoost for Optimized Job Allocation

recommendation-on-demand, where contextual

information and listening habits are indigenously

considered while making recommendations. Future

research will focus on enhancing the system’s

responsiveness to evolving user preferences and fine-

tuning recommendation algorithms for different user

categories.

Reddy et al. (Reddy et al., 2023) present the

challenge of predicting flight delays, especially those

induced by bad weather. Having trained various

machine learning algorithms on an integrated dataset

of weather and flight results from JFK airport, they

determined that XGBoost performed best, achieving

an RMSE with a severity of 0.81. The current

obstacles remain improving the model’s

responsiveness to real-time data and addressing other

factors influencing flight delays.

Pecheti et al. (Pecheti et al., 2024) present the

Drug Information and Recommendation System that

draws on Amazon Web Services (AWS) to support

drug review opinions. Their work shows a design

approach involving data collection, preprocessing,

and prediction on drugs ultimately leading to the

deployment of the Sentiment AI platform. The system

ensures both scalability and software security by

utilizing services provided by AWS such as EC2 and

S3 and IAM. Future improvements are intended to

lead an expansion of data sources regarding the

system and improvement of analytical capacity

within the real-world environment.

Reddy et al. (Reddy et al., 2024), proposed a

sentiment analysis model based on Long Short-Term

Memory (LSTM) and natural language processing

algorithms for evaluating user reactions to YouTube

content. Trained on the IMDB dataset alongside

AWS, the model provides an avenue for further

enhancements to widen the dataset for better

generalization whilst working on the interactive

dashboard to aid forward an even deeper user insight.

Selvi, S. et al. (Selvi and Manimegalai, 2024)

Proposed new optimization techniques [Multiverse

Optimization (MVO)], which enhances the efficiency

for task scheduling taking advantage of neighborhood

structures. This approach brings several benefits,

including significantly reduced energy consumption

and degradation of makespan as it can be verified

through laboratory testbed results with improved

performance metrics in contrast to other scheduling

approaches, which we have outlined above. This

study demonstrates that the proposed scheduler will

be able, through experimental evaluation, to schedule

tasks appropriately resulting in minimization of

operational costs for a green cloud computing

environment. Although the results are impressive,

more research is needed to solve these scalability

challenges. There is much more work to do in this

area and there are no clear answers yet on how

effective resource management should be for the

future of cloud computing.

2.2 Research Gap

An integrated approach shows a huge research gap in

pressure, temperature, and water flow sensors’

behavior among racks in data centers while they are

in operation. Even though behavior of individual

sensors is researched on its own, the application of

machine learning models in supporting prediction and

understanding of the collective behavior of multi-

sensor usage in dynamic and intense workload

environments is not clear. Also, another question to

be answered is the training of machine learning

algorithms for analyzing anomalies or inefficiencies

across different sensors, which are revealed through

discrepancies from normal behavior. Of course, these

projects will make data center operations better,

resource usage optimization, and real-time

monitoring more accurate.

3 METHODOLOGY

This paper presents a clear framework for the

systematic creation of predictive models from sensor

data. It is the data transformation, developing

machine learning models, benchmarking and fine

tuning of the models and the solutions ready for

implementation. It goes through the steps of data

acquisition, data preparation, data transformation and

feature extraction, model building, model assessment

and, model refinement, and concerns of model

deployment. Here’s a detailed breakdown of the steps

involved.

3.1 Data Acquisition and Analysis

The raw data consists of data measured by sensors,

and includes data from 20 racks where each rack had

seventy-two servers. In every server employed, job

execution finds 24 cores at its behest. The dataset

includes readings from eleven sensors for each rack,

monitoring various parameters: The following is a list

of links status, rackcdu liquid level, rackcdu pressure,

facility pressure, facility water flow, rackcdu leak

detection, heat load (sampling rate of 60s),facility

water temperature supply, facility water temperature

return, server liquid temperature supply, and server

liquid temperature return. Some of the recognized

ISPES 2024 - International Conference on Intelligent and Sustainable Power and Energy Systems

attributes include; Device, Sensor, Time, Value, and

Units where measurements include pressure,

temperature, and Water flow parameters.

3.2 Preprocessing

This means that preprocessing helps with the quality

of the data that will be fed into a machine learning

algorithm as well as how consistent that data is. The

units are normalized against each other in accordance

with each sensor’s specific type, where each type is

assigned its index. It is in this normalization that

uniformities as well as accurate training models are

made possible. The case when a have missing or

inconsistent values of data points is a typical problem

of preprocessing; such methods as interpolation or

data imputation are applied to complete the gaps as

well as to manage outliers, thus preparing a suitable

data set for the model.

3.3 Feature Selection

Feature engineering is more important, where all

these features are chosen and structured or formatted

in such a way that would take good results on the

predictive models. In this regard, the analysis of the

sensor data leads to the mapping of the values to

particular labels including the facility water flow or

server liquid temperature of the system which serves

as the features in the training model phase.

Feature selection is performed according to the

measures’ importance and relevance to the target

variable, for example, for heat load or water

temperature prediction. There is no doubt that having

domain knowledge is very important in selecting the

most significant features. Furthermore, there could be

literals conducted to achieve new features that are

more suitable for revealing the interdependencies

inside the data to improve the model’s predictability.

In this study, two features were selected. They are:

• Units: The Type of Measurement of sensors in the

datacentres for a particular period of time.

• Value: The Result of measurement of sensors in the

datacentres for a particular period of time.

3.4 Model Training and Evaluation

For the predictive modeling step, the machine

learning algorithms are used, such as regression

models: Random Forest and XGBoost. Both

algorithms are a kind of learning algorithms that are

used to forecast target variable based on the input of

characteristic sensors. Hyperparameter tuning is then

done in each model for better results, all of them have

been trained on the pre-processed and the features

that were engineered.

Before defining the hyperparameters, they have to be

tuned properly by using the grid search or randomized

search, which ensures higher accuracy and model

generalization. Case of Random Forest, other

parameters like number of estimators and tree depth

are tuned while for XGBoost the boosting parameters

such as learning rate and maximum tree depth are

tuned.

The effectiveness of the models is estimated using

Mean Squared Error (MSE), Root Mean Squared

Figure 1: Model Architecture

Error (RMSE), Mean Absolute Error (MAE), and the

Coefficient of Determination (R2). These are

information on the performance of the model and it

capability in predicting on unseen data.

Cross-validation is used to overcome the problem of

overfitting so that inherited property prediction is

made with high reliability. It attempts to divide the

data into the training and the validation part many

times, with each new model acting on a different

division. This approach offers a complete evaluation

of the different model performance on different data

samples and the necessary adjustments are made.

Optimization of a model goes further from the given

training to improve the prediction capability and

reduce computation time. Algorithm-specific

optimizations are the following ones: for example, in

case of Random Forest, the importance of features is

used to prune less important features. In the case of

XGBoost, there are tuning parameters like the

number of iterations boosting the model, learning

rate, and maximum depth are set in detail to optimize

the model’s efficiency.

3.5 Overview

The below Fig. 1 outlines a comprehensive machine

learning workflow, starting with Data Acquisition to

gather relevant data, followed by Filtering Data to

clean and preprocess it. The process continues with

Data Splitting to create training and testing sets, then

moves to Training and Evaluation of models, often

incorporating Cross-Validation to optimize

performance. After training, Visualization of Results

provides insights through graphical representation,

Analyzing Facility Servers Using Random Forest and XGBoost for Optimized Job Allocation

followed by Model Comparison to identify the best-

performing model. The results are then Saved and

Exported, and the workflow concludes with

Deployment, where the refined model is implemented

for real-world use.

4 RESULTS AND ANALYSIS

Table 1 also demonstrates the performance of the

model Random Forest Regressor using Mean

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

Root Mean Squared Error (RMSE) and R-squared

(R2) of twenty different racks. Both 70/30 and 80/20

splits are provided also with/without cross-validation.

This is demonstrating how cross-validation affects

the prediction by having some cases with increased

R2 than others with a worse performance when cross-

validation is applied. Rear observations are high R2

values-the values literally near or more than 0.95 for

most racks without cross validating and cross

validating has shown much less values that can be

good indicators of overfitting racks. Moreover,

improvements are noticed when cross-validation is

incorporated in some racks especially Rack 16 whose

R2 rises to as high as 0.445 in the loop.

Table 1: Rack Wise Results for RF Regressor only for

Numeric Values

Rack

no.

Split CV MAE MSE R2

Rack 1

70/30 No 58.621 64189.336 0.996

70/30 Yes 2917.325 2756325.59 -0.836

80/20 No 58.472 64661.218 0.996

80/20 Yes 2917.325 2756325.59 -0.836

Rack 2

70/30 No 56.3 48833.29 0.994

70/30 Yes 560.939 1273912.904 0.948

80/20 No 56.39 49663.522 0.993

80/20 Yes 560.939 1273912.904 0.948

Rack 3

70/30 No 35.682 25434.312 0.989

70/30 Yes 699.939 3608070.746 0.69

80/20 No 34.947 23347.428 0.99

80/20 Yes 699.939 3608070.746 0.69

Rack 4

70/30 No 65.503 111249.705 0.973

70/30 Yes 865.975 4039189.745 0.73

80/20 No 62.672 110123.774 0.974

80/20 Yes 865.975 4039189.745 0.73

Rack 5

70/30 No 150.078 439327.808 0.875

70/30 Yes 1656.948 18500900.81 -0.16

80/20 No 49.89 42524.393 0.997

80/20 Yes 1656.948 18500900.81 -0.16

Rack 6

70/30 No 49.423 38197.812 0.992

70/30 Yes 1371.369 5677594.478 -0.143

80/20 No 49.065 38118.672 0.992

80/20 Yes 1371.369 5677594.478 -0.143

70/30 No 33.622 39029.145 0.952

Rack 7

70/30 Yes 1394.243 7975824.019 0.018

80/20 No 34.002 16941.352 0.998

80/20 Yes 1394.243 7975824.019 0.018

Rack 8

70/30 No 35.233 20647.193 0.997

70/30 Yes 1394.243 7975824.019 0.018

80/20 No 34.002 16941.352 0.998

80/20 Yes 1394.243 7975824.019 0.018

Rack 9

70/30 No 113.636 260977.59 0.982

70/30 Yes 1611.013 17412938.19 -0.18

80/20 No 114.719 268206.437 0.982

80/20 Yes 1611.013 17412938.19 -0.18

Rack 10

70/30 No 190.414 433605.361 0.943

70/30 Yes 1719.228 8069397.218 -0.141

80/20 No 185.795 423796.024 0.878

80/20 Yes 1719.228 8069397.218 -0.141

Table 2 illustrates the performance of the XG-Boost

Regressor model on 20 different racks with numeric

test data based on MAE, MSE, RMSE, and R2. We

report results for four splits 70/30 and 80/20 with and

without CV. In general, the R2 values are high (above

0.9) though it does not apply cross-validation but

when applying cross-validation, the vast racks such

as Rack 1 that originally had an R2 of 0.147,

drastically drop hugely to an R2 if -1.643. Even more,

some racks, namely Rack 5 and Rack 6, also

demonstrate the decrease in the value of R2 after the

cross validation, which also points to the overtraining

of models. However, some racks, for instance, Rack

16 have a positive R2 of 0.992 without cross-

validation but a negative value with cross-validation.

The results emerge in terms of the inconsistency of

utilising cross-validation in the XGBoost model,

where the variation of R2 values and errors is large

among different racks.

Table 2: Rack Wise Results using XGBoost Regressor for

Numeric Values Test Data 10 Racks

Rack

No.

Split CV MAE MSE R2

Rack 1

70/30 No 66.783 82676.425 0.994

70/30 Yes 3536.528 39687716.82 -1.643

80/20 No 66.891 86840.584 0.994

80/20 Yes 3536.528 39687716.82 -1.643

Rack 2

70/30 No 70.161 102338.714 0.995

70/30 Yes 576.594 1422092.694 0.943

80/20 No 69.154 98179.045 0.995

80/20 Yes 576.594 1422092.694 0.943

Rack 3

70/30 No 63.298 106335.139 0.991

70/30 Yes 698.085 3545491.34 0.691

80/20 No 62.871 107386.452 0.991

80/20 Yes 698.085 3545491.34 0.691

Rack 4

70/30 No 70.562 157762.494 0.99

70/30 Yes 890.744 14535632.96 -0.607

80/20 No 127.833 1638371.263 0.928

ISPES 2024 - International Conference on Intelligent and Sustainable Power and Energy Systems

80/20 Yes 1530.762 2014891.83 -0.212

Rack 5

70/30 No 72.424 104194.679 0.993

70/30 Yes 1655.008 18499826.64 -0.16

80/20 No 71.682 102993.421 0.994

80/20 Yes 1655.008 18499826.64 -0.16

Rack 6

70/30 No 60.03 75719.105 0.985

70/30 Yes 1705.83 8809078.817 0.069

80/20 No 257.811 414063.18 0.93

80/20 Yes 1653.531 9372708.78 0.079

Rack 7

70/30 No 46.446 62130.995 0.986

70/30 Yes 1885.324 16551543.82 -1.038

80/20 No 46.645 55290.557 0.993

80/20 Yes 1885.324 16551543.82 -1.038

Rack 8

70/30 No 70.711 148246.431 0.992

70/30 Yes 2133.422 26405897.6 -0.515

80/20 No 70.766 149922.959 0.992

80/20 Yes 2133.422 26405897.6 -0.515

Rack 9

70/30 No 134.751 408264.553 0.976

70/30 Yes 1639.415 17501719.73 -0.186

80/20 No 136.184 4068378.53 0.976

80/20 Yes 1639.415 17501719.73 -0.186

Rack 10

70/30 No 117.404 308557.061 0.987

70/30 Yes 1540.57 18859281.3 -0.141

80/20 No 115.255 303458.405 0.982

80/20 Yes 1540.57 18859281.3 -0.141

Comparison of Table 1 and Table 2 show that the RF

Regressor has effectively learned from the data and

consistently performs well across rakes and data

splits, as evidenced by the higher R² values, often

approaching or exceeding 0.9; thereby confirming a

strong correlation between predicted and actual

values. For example, in Rack 1 with a 70/30 split and

no cross-validation, the RF Regressor records 0.996

R², while XGBoost gets 0.994. As against the

performance of the RF Regressor, the XGBoost

Regressor shows massive variability in performance-

and wades through the data it learns awfully even

under similar conditions. Many of the R² values are

negative or very close to 0-in particular, for the 80/20

splits-suggesting that XGBoost does not capture the

underlying patterns well. For example, Rack 10’s R²

was -0.141 at an 80/20 split with cross-validation for

XGBoost, while the same scenario for RF Regressor

resulted in 0.968.

The two graphs shown in Fig. 2 (a), (b) presents the

predicted vs. true values of a Random Forest model,

its performance with and without cross- validation is

highlighted. In the first graph Points are blue and

scattered but their corresponding points seem to lie

near the ideal fit line but not completely perfect as

some of them are a little farther, this is because of

overfitting but not a serious one. In comparison, the

second graph where the residuals were corrected with

cross-validation presents green points away from the

ideal fit line more often and specially at higher values,

which suggests lower reliability and stochasticity in

the validation folds.

(a) Rack 4 in 70:30 Split ratio

(b) Rack 4 in 80:20 split Ratio

Figure 2: Scatter Plot between True Values and Predicted

Values in Random Forest Regressor for Rack 4

(a)

Rack 10 in 70:30 Split ratio

(b) Rack 10 in 80:20 split Ratio

Figure 3: Scatter Plot between True Values and Predicted

Values in Random Forest Regressor for Rack 10

This comparison shows the model’s performance

to cross-validation and the difficulty of maintaining a

stable level of predictive accuracy across different

data splits.

Analyzing Facility Servers Using Random Forest and XGBoost for Optimized Job Allocation

(a) Rack 3 in 70:30 Split ratio

Rack 3 in 80:20 split Ratio

Figure 4: Scatter Plot between True Values and Predicted

Values in XGBoost Regressor for Rack 3

Rack 6 in 70:30 Split ratio

Rack 6 in 80:20 split Ratio

Figure 5: Scatter Plot between True Values and Predicted

Values in XGBoost Regressor for Rack 6

The two plots above Fig. 5 (a), (b) illustrates the

comparison between the XGBoost model

performance with the model which include the cross

Validation for Rack 6. The left plot shows the results

when cross-validation is not done while the

predictions are depicted using blue circles. The

correct plot includes cross-validation check, and

predictions are marked in green dots. PredPol: In both

the plots above, the dotted line line indicates the

Which indicates the true positive or perfect fit line

that equates the true values to the predicted values.

Also, when making predictions without using cross-

validation they seem to be distributed farther and are

a less accurate representation of the ideal line because

of this, the line on the right shows how predictions

with cross-validation look like and demonstrate how

cross-validation affects the consistency and ability to

generalize when making predictions. This

comparative analysis discusses the effect of cross-

validation on the result of the model, in terms of

accuracy and behavior.

5 CONCLUSIONS

This research shows how a Random Forest and XG-

Boost can be used to identify outliers in a data centres.

From regression tasks in these models, meaning- ful

information regarding the performance of different

types of racks was obtained. More work in the future

will be towards the analysis and prediction of patterns

of job assignments using the sensors. Moreover, the

use of these models to create simulation environments

should be the focus of future research because it will

allow better control over data center management.

ACKNOWLEDGEMENTS

Sincere gratitude is expressed to the university,

Amrita Vishwa Vidyapeetham and Dr. Beena B. M.,

for their assistance in writing the paper.

REFERENCES

Abbas, G., Hatatah, M., Ali, A., Touti, E., Alshahir, A., and

Elrashidi, A. M. (2023). A novel energy proficient

computing framework for green computing using

sustainable energy sources. IEEE Access, 11:126542–

126554.

Ahmad, A., Khan, S. U., Khan, H. U., Khan, G. M., and

Ilyas, M. (2021). Challenges and practices

identification via a systematic literature review in the

adoption of green cloud computing: client’s side

approach. IEEE Access, 9:81828– 81840.

Beloglazov, A. and Buyya, R. (2014). An energy- aware

combinatorial virtual machine allocation and placement

model for green cloud computing. In Proceedings of the

International Conference on Cloud Computing and

Technology (Cloud- Com), pages 1–8, Sydney,

Australia.

Chiaraviglio, L., Mellia, M., and Neri, F. (2014). Eco-aware

online power management and load scheduling for

green cloud datacenters. IEEE Transactions on

Networking, 22(1):1–14.

ISPES 2024 - International Conference on Intelligent and Sustainable Power and Energy Systems

Guo, Z., Li, J., and Ramesh, R. (2023). Green data analytics

of supercomputing from massive sensor networks:

Does workload distribution matter? Information

Systems Research, 34(4):1664–

1685.

Kulkarni, V. V., Vishal, S., and Beena, B. (2024). Cloud

based music recommendation system. Available at

SSRN 4814206.

Pecheti, S. T., Kodurupaka, N., Basavadeepthi, H. M.,

Tanvi, T., and Beena, B. M. (2024). Leveraging cloud

computing for drug review analysis. In SCI, pages 150–

160.

Qiu, C., Shen, H., and Chen, L. (2018). Towards green

cloud computing: Demand allocation and

pricing policies for cloud service brokerage.

IEEE Transactions on Big Data, 5(2):238–251. Qiu, M.,

Ming, Z., Li, J., Gai, K., and Zong, Z.

(2015). Phase-change memory optimization for

green cloud with genetic algorithm. IEEE Transactions on

Computers, 64(12):3528–3540.

Raja, S. P. (2021). Green computing and carbon footprint

management in the it sectors. IEEE Transactions on

Computational Social Systems, 8(5):1172–1177.

Reddy, P. N., Aswath, S. S., Alapati, R., and Beena,

B. M. (2024). Aws enhanced sentiment analysis using lstm

for online video comments. In SCI, pages 193–205.

Reddy, R. T., Pati, P. B., Deepa, K., and Sangeetha,

S. T. (2023). Flight delay prediction using machine

learning. In IEEE 8th International Conference for

Convergence in Technology (I2CT), Lonavla, India.

Selin, N. E. (2024). Carbon footprint.

Encyclopaedia Britannica. Available:

https://www.britannica.com/science/carbon- footprint.

Selvi, S. and Manimegalai, D. (2024). A neigh- borhood

inspired multi-verse scheduler for energy and

makespan optimized task scheduling for green cloud

computing systems. International Journal of

Information and Computer Security, 16(2):123–145.

Skourletopoulos, G., Mavromoustakis, C. X., Mastorakis,

G., Batalla, J. M., Song, H., Sahalos,

J. N., and Pallis, E. (2018). Elasticity debt analytics

exploitation for green mobile cloud computing: An

equilibrium model. IEEE Transactions on Green

Communications and Networking, 3(1):122–131.

Xiao, Y. and Li, Z. (2018). Energy-efficient hybrid

framework for green cloud computing. IEEE

Transactions on Green Communications and

Networking, 2(2):202–212.

Analyzing Facility Servers Using Random Forest and XGBoost for Optimized Job Allocation