TELKOMNIKA Telecommunication Computing Electronics and Control

Vol. 23, No. 3, June 2025, pp. 694~702

ISSN: 1693-6930, DOI: 10.12928/TELKOMNIKA.v23i3.26554  694

XGBoost optimization using hybrid Bayesian optimization and

nested cross validation for calorie prediction

Budiman1, Nur Alamsyah2, Titan Parama Yoga2, R. Yadi Rakhman Alamsyah2, Elia Setiana1
1
Department of Informatics Engineering, Faculty of Information and Technology, Universitas Informatika dan Bisnis Indonesia,
Bandung, Indonesia
2
Department of Information System, Faculty of Information and Technology, Universitas Informatika dan Bisnis Indonesia, Bandung,
Indonesia

Article Info ABSTRACT

Article history: Accurately predicting calorie expenditure is crucial for wearable device
applications, enabling personalized fitness and health recommendations.
Received Aug 9, 2024 However, traditional models struggle with high data variability and
Revised Jan 23, 2025 nonlinear relationships in activity data, leading to suboptimal predictions.
Accepted Mar 23, 2025 This study addresses these challenges by integrating extreme gradient
boosting (XGBoost) with Bayesian optimization and nested cross validation
to enhance predictive accuracy. Unlike previous approaches, our method
Keywords: systematically tunes hyperparameters using Bayesian optimization while
employing nested cross validation to prevent overfitting, ensuring robust
Bayesian optimization model evaluation. We utilize a dataset of daily activity records, including
Calorie prediction steps, distance, and active minutes, extracted from wearable devices. Our
Extreme gradient boosting experimental findings indicate a substantial enhancement in prediction
Nested cross validation performance, achieving a mean squared error (MSE) of 4294.27, an R-
Wearable devices data squared (R2 ) score of 0.9917, and a root mean squared error (RMSE) of
65.53. The proposed model outperforms baseline approaches such as random
forest and support vector machines in terms of predictive accuracy. These
findings underscore the advantage of our approach in predictive modeling.
Beyond calorie estimation, the proposed methodology is adaptable to other
domains requiring high-precision predictions, such as healthcare analytics
and personalized recommendation systems.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Nur Alamsyah
Department of Information System, Faculty of Information and Technology
Universitas Informatika dan Bisnis Indonesia
St. Soekarno Hatta, Bandung, Indonesia
Email: [email protected]

1. INTRODUCTION
The rapid advancement of wearable technology has transformed the health and fitness industry,
enabling continuous real-time tracking of physical activity. Devices such as smartwatches and fitness trackers
collect extensive data, including steps taken, distance traveled, active minutes, and calorie expenditure [1],
[2]. Accurate calorie prediction is essential for personalized fitness recommendations, weight management,
and preventive healthcare [3], [4]. Accurately predicting calorie expenditure is crucial for personalized
fitness recommendations, weight manage- ment, and overall health improvement [5], [6]. Despite the
accessibility of large amounts of data, developing robust models to predict daily caloric expenditure remains
a challenging task due to the complex interactions of various factors that influence energy spending. A
prominent issue in predicting calorie expenditure is the need for accurate and reliable models that can handle

Journal homepage: http://journal.uad.ac.id/index.php/TELKOMNIKA

TELKOMNIKA Telecommun Comput El Control  695

the high dimensionality and variability of the data [7]. Traditional methods, such as linear regression, often
fail to capture the nonlinear relationships that are inher- ent in wearable data. More recent advancements in
machine learning, particularly ensemble methods such as extreme gradient boosting (XGBoost), have shown
promise in addressing this challenge [8]. However, the performance of these models relies heavily on precise
hyperparameter tuning, which can be computationally expensive and vulnerable to overfitting [9], [10].
The literature on calorie prediction using machine learning has highlighted various approaches that
have been applied. For instance, previous studies have used random forests and support vector machines,
which have shown moderate success in calorie prediction [11]. One study by Hwang et al. [12] used random
forests to predict calorie expenditure based on daily activity data, achieving an R-squared (R2 ) score of 0.82.
However, these methods often lack the robustness and predictive accuracy required for practical applications.
In addition, another study by Lee [13] used linear regression and neural networks for calorie prediction, but
faced similar challenges related to non-linear relationships and the complexity of data from wearable devices.
This study showed that neural networks can provide better predictions than linear regression, but are still
limited in handling high data variation [14]. Although XGBoost has been applied in various domains with
significant success, its application in calorie prediction coupled with advanced hyperparameter optimization
techniques is still rarely explored [15].
Research by Aziz et al. [16] showed that XGBoost has great potential in calorie prediction,
achieving an R2 score of 0.87 when applied to physical activity data. However, this study has not utilized
more advanced hyperparameter optimization such as Bayesian optimization. These studies generally show
that there is great potential in improving prediction accuracy through more advanced and integrated
approaches. In this study, our main contribution is to propose a novel approach that leverages XGBoost with
Bayesian optimization and nested cross validation to improve the accuracy of calorie expenditure prediction.
Bayesian optimization systematically searches the hyperparameter space, finding optimal values more
efficiently than traditional grid search methods. Nested cross validation further ensures that model
performance is robustly validated, reducing the risk of overfitting and providing reliable estimates of model
accuracy. To address these gaps, this study makes the following key contributions: i) utilization of XGBoost
for calorie prediction, leveraging its ability to model nonlinear relationships effectively; ii) implementation of
Bayesian optimization for hyperparameter tuning, systematically searching the hyperparameter space to
enhance model efficiency; and iii) application of nested cross validation to improve model generalization and
prevent overfitting, ensuring a robust evaluation framework.
Our proposed approach combines XGBoost with Bayesian optimization and nested cross validation,
offering an innovative solution for high-accuracy calorie prediction from wearable data. By optimizing
hyperparameters efficiently and incorporating a robust validation strategy, this study outperforms
conventional methods and provides a more reliable predictive framework. The findings contribute to
personalized health monitoring and can be extended to other domains requiring precise predictive modelling.

2. METHOD
The method used in this study is to predict calories based on daily activity data collected using Fitbit
devices. The method consists of several main stages, namely data collection, data preprocessing, data split-
ting, model definition, hyperparameter tuning (using GridSearchCV and nested cross validation), and model
evaluation. Our proposed method is presented in Figure 1.

Figure 1. Proposed method

XGBoost optimization using hybrid Bayesian optimization and nested cross validation for … (Budiman)
696  ISSN: 1693-6930

2.1. Data collected

The data used in this study was obtained from the Kaggle portal with the topic “Fitbit fitness tracker
data: capstone project”. This dataset contains users’ daily activity records collected using Fitbit devices. The
dataset consists of 940 rows and 15 columns that include various features such as number of steps, distance
traveled, active minutes, and calories burned. This data provides comprehensive information about the user’s
daily physical activity, which is then used to predict calorie expenditure. A description of the features in the
data used in this study is presented in Table 1.

Table 1. Feature descriptions for fitbit fitness tracker data

Variable Description (reworded)
Id A unique identifier assigned to each individual user.
ActivityDate The specific date on which the recorded activity took place.
TotalSteps The cumulative count of steps taken throughout the day.
TotalDistance The total miles travelled within a single day.
TrackerDistance The distance measured and logged by the tracking device.
LoggedActivitiesDistance The distance covered during explicitly recorded activities.
VeryActiveDistance The distance accumulated while performing highly vigorous activities.
ModeratelyActiveDistance The total distance travelled during moderately intense movements.
LightActiveDistance The measured distance from light-intensity activities.
SedentaryActiveDistance The distance recorded while engaging in low-movement or stationary activities.
VeryActiveMinutes The total time, in minutes, spent in high-intensity physical exertion.
FairlyActiveMinutes The sum of minutes spent in moderately intense activities.
LightlyActiveMinutes The duration of low-effort physical movement throughout the day.
SedentaryMinutes The total amount of time spent in an inactive or resting state.
Calories The total number of calories expended over the course of the day.

2.2. Data preprocessing

The data preprocessing stage is conducted to maintain data quality and ensure its readiness for
model training [17]. The initial step involves addressing missing values, either by imputing them or
discarding incomplete records. Subsequently, non-essential features, such as ActivityDate, are eliminated to
prevent analytical inconsistencies. The dataset is then subjected to normalization or standardization, aligning
all features to a uniform scale, which is particularly crucial for specific machine learning algorithms.
Moreover, if categorical variables are present, they are transformed into numerical representations to
facilitate model processing. The outcome of this preprocessing phase is a well-structured dataset, optimized
for training and evaluation purposes.

2.3. Data splitting

After completing the data preprocessing stage, the dataset is then partitioned into two subsets: a
training set and a testing set. This segmentation ensures that the model is evaluated on previously unseen
data, allowing for a more reliable assessment of its performance [18]. In this study, the dataset was randomly
split, with 80% allocated for training and 20% for testing, ensuring a well-balanced distribution. The training
set is utilized to develop the model, while the testing set serves as an independent benchmark to measure its
predictive accuracy.

2.4. Model definition

At this phase, the machine learning model to be used for the calorie prediction is defined. In this study,
we used XGBoost, a powerful and efficient ensemble learning algorithm [19], [20]. XGBoost was chosen for
its capabilities in handling complex data and producing accurate predictions with high computational
speed [21]. The model is able to capture non-linear relationships in the data and automatically handle missing
features, making it a suitable choice for calorie prediction based on physical activity data collected from Fitbit
devices [22]. The basic parameters of the XGBoost model are set first, which will then be optimized in the next
stage of hyperparameter tuning [23]. XGBoost is an ensemble-based machine learning algorithm that
leverages decision trees as its foundational learning units [24]. It optimizes an objective function consisting
of a loss function and a regulation term to prevent overfitting. The optimization function for XGBoost can be
formulated as (1).

ℒ(θ) = ∑𝑛𝑖=1 𝑙(𝑦̂,𝑖 𝑦𝑖 ) + ∑𝐾

𝑘=1 Ω(𝑓𝑘 ) (1)

The function (ℒ(θ)) represents the optimization objective, while (l(ŷi , yi )) serves as the loss
function, quantifying the discrepancy between the predicted value (ŷ) i and the actual value (yi ).

TELKOMNIKA Telecommun Comput El Control, Vol. 23, No. 3, June 2025: 694-702
TELKOMNIKA Telecommun Comput El Control  697

Additionally, (Ω(fk )) acts as the regularization component to mitigate overfitting. In regression tasks, one of
the most commonly employed loss functions is the mean squared error (MSE), mathematically expressed
as (l(ŷ,i yi ) = (ŷi − yi )2 ). The regularization term (Ω(fk )) is generally defined as (2).
1
(Ω(𝑓𝑘 ) = γ𝑇 + λ ∑𝑇𝑗=1 𝑤𝑗2 ) (2)
2

(γ) and (λ) function as hyperparameters for regularization, while (T) denotes the total number of
leaves in the decision tree. Additionally, (wj ) represents the weight assigned to the (j)-th leaf. Consequently,
the complete optimization function for XGBoost in the context of regression with regularization can be
expressed as (3).

1
(ℒ(θ) = ∑𝑛𝑖=1(𝑦̂𝑖 − 𝑦𝑖 )2 + ∑𝐾 𝑇 2
𝑘=1 (γ𝑇 + λ ∑𝑗=1 𝑤𝑗 ) (3)
2

The XGBoost model aims to minimize this objective function during training to produce an optimal
model. This approach allows XGBoost to handle complex data and provide accurate predictions with high
computa- tional efficiency, making it a suitable choice for predicting calorie expenditure based on physical
activity datacollected from Fitbit devices.

2.5. Hyperparameter tuning (GridSearchCV+nested cross validation)

To optimize the performance of the XGBoost model, hyperparameter tuning is carried out using
GridSearchCV in combination with nested cross validation [25]. The purpose of hyperparameter tuning is to
determine the best parameter set that minimizes the objective function, thereby enhancing both accuracy and
robustness. GridSearchCV conducts a comprehensive search across a predefined parameter grid, training the
model and assessing its performance through cross-validation for each parameter combination [26]. The most
effective set of hyperparameters is then chosen based on the cross-validation results.
In this study, key hyperparameters that were fine-tuned include the learning rate (η), the maximum
tree depth (max depth), and the number of trees (n estimators). To further improve model reliability and
mitigate overfitting, nested cross validation is implemented. This method consists of two cross-validation
loops: an outer loop dedicated to model evaluation and an inner loop for hyperparameter optimization. By
employing this approach, an unbiased assessment of the model’s performance is obtained. The optimization
function for XGBoost, incorporating hyperparameter tuning, is expressed as (4).

ℒ(θ) = ∑𝑛𝑖=1 𝑙(𝑦̂,𝑖 𝑦𝑖 ) + ∑𝐾

𝑘=1 Ω(𝑓𝑘 ) (4)

Where the loss function (l(ŷi , yi )) is defined as (5):

𝑙(𝑦̂,𝑖 𝑦𝑖 ) = (𝑦̂𝑖 − 𝑦𝑖 )2 (5)

and the regularization term (Ω(fk )) is given by (6).

1
Ω(fk ) = γT + λ ∑Tj=1 wj2 (6)
2

By minimizing this objective function, the hyperparameter tuning process aims to find the optimal
values for (η), (max_depth), and (n_estimators), resulting in a model that generalizes well to new data.

2.6. Evaluation
After training and optimizing the XGBoost model using GridSearchCV and nested cross validation,
the next phase involved assessing its performance. The evaluation was conducted using a pre-designated test
dataset to verify the model’s ability to accurately predict calorie expenditure on previously unseen data.
Several essential metrics were utilized for performance assessment, including MSE, R2 score, and root mean
squared error (RMSE). These evaluation measures help determine the model’s effectiveness in calorie
prediction, ensuring high accuracy and strong generalization to new data. The results of the evaluation
indicate that the proposed approach outperforms existing methods in calorie prediction based on physical
activity data.

XGBoost optimization using hybrid Bayesian optimization and nested cross validation for … (Budiman)
698  ISSN: 1693-6930

3. RESULTS AND DISCUSSION

The performance evaluation of the proposed model is conducted using various metrics, including
MSE, R2 score, and RMSE. These metrics offer a more thorough analysis of the model’s precision and its
capability to adapt to unseen data.

3.1. Result
The performance of different models was assessed using MSE, R2 score, and RMSE. A summary of
the evaluation results is presented in Figure 2. The random forest model, without hyperparameter tuning,
produced an MSE of 69,109.31, an R2 score of 0.86, and an RMSE of 262.89. This baseline model exhibited
moderate performance but had potential for enhancement. The XGBoost model, optimized using
GridSearchCV, demonstrated notable improvements, achieving an MSE of 50562.42, an R 2 score of 0.90,
and an RMSE of 224.86. This indicates that XGBoost, even with basic hyperparameter tuning, outperforms
the random forest model.

Figure 2. Evaluation result

Further enhancement was achieved with XGBoost combined with Bayesian optimization, resulting
in an MSE of 44120.17, an R2 score of 0.91, and an RMSE of 210.05. This demonstrates the effectiveness
of Bayesian Optimization in improving the model’s predictive accuracy. When repeated cross validation was
added to the XGBoost model tuned with Bayesian optimization, the performance metrics improved
sub-substantially, with an MSE of 11049.92, an R2 score of 0.98, and an RMSE of 105.12. This approach
shows a significant reduction in error and an increase in model reliability. The best results were obtained
with XG-Boost using Bayesian optimization and nested cross validation. This model achieved an MSE of
4294.27, anR2 score of 0.99, and an RMSE of 65.53. The combination of these advanced techniques provided
the most accurate and robust model, highlighting their importance in model tuning and validation.

TELKOMNIKA Telecommun Comput El Control, Vol. 23, No. 3, June 2025: 694-702
TELKOMNIKA Telecommun Comput El Control  699

3.2. Discussion
The results of this study demonstrate significant improvements in predicting calorie expenditure using
XGBoost with Bayesian optimization and nested cross validation. Our final model achieves superior per-
formance, evidenced by the highest R2 score of 0.99 and the lowest RMSE of 65.53. Comparing these results
to similar studies highlights the advancements made in our ap- proach. For instance, Elshewey et al. [27]
employed Random Forest and reported an R2 score of 0.85 and an RMSE of 200. Their study, while
comprehensive, did not integrate advanced hyperparameter tuning methods like Bayesian Optimization,
which we found crucial for enhancing model performance. Another study by Wu [14] used support vector
machines and achieved an R2 score of 0.88. Although their model pro- vided reasonable predictions, it lacked
the robustness and accuracy observed in our XGBoost model optimized with nested cross validation.
Further, a recent study by Asadi and Hajj [28] using neural networks achieved an R2 score of
0.91 and an RMSE of 180. Their approach, while effective, underscores the potential of ensemble methods
like XGBoost, especially when combined with sophisticated tuning techniques. Similarly, a study published
in the International Research Journal of Modernization in Engineering Technology and Science (2023)
highlighted the effectiveness of machine learning models, reporting an R 2 score of 0.93 using gradient
boosting techniques, but without employing nested cross-validation for model validation, which might have
resulted in a less reliableperformance estimation.
The novelty of our research lies in integrating Bayesian optimization with nested cross validation to
develop an effective predictive model for calorie expenditure. Our experimental results indicate notable
enhancements in prediction accuracy, achieving a MSE of 4294.27, an R2 score of 0.9917, and RMSE of
65.53. These results highlight the potential of our approach in delivering more precise and dependable calorie
predictions, which are essential for personalized fitness and health management.

4. CONCLUSION
This study successfully developed an advanced predictive model for calorie expenditure estimation
using XGBoost with Bayesian optimization and nested cross validation. The proposed approach significantly
improves prediction accuracy and robustness, as demonstrated by the final model achieving an R² score of
0.99 and an RMSE of 65.53, outperforming baseline models. These findings highlight the effectiveness of
integrating hyperparameter tuning and robust validation techniques in predictive modeling. Beyond
theoretical advancements, the proposed model has practical applications in real-time fitness tracking apps and
health monitoring platforms. By providing more accurate calorie expenditure estimates, this model can
enhance personalized fitness recommendations, support weight management programs, and assist healthcare
providers in monitoring physical activity levels.
Despite its promising results, this study has certain limitations. First, the model relies on structured
wearable data, which may not fully capture dynamic real-world variations in physical activity. Future
research can explore deep learning approaches, such as long short-term memory (LSTM) or convolutional
neural network (CNN) models, to handle more complex temporal patterns in calorie expenditure.
Additionally, incorporating real-time data streams from internet of things (IoT)-enabled wearables could
further improve prediction reliability. Another potential direction is applying this approach to other health
metrics, such as heart rate variability, sleep tracking, or stress levels, expanding its utility beyond calorie
estimation. Overall, the study demonstrates that advanced machine learning techniques can significantly
enhance predictive accuracy in health and fitness applications. By leveraging continuously growing datasets
from wearable technology, future research can refine these models to develop even more accurate, real-time,
and personalized health monitoring systems.

ACKNOWLEDGMENTS
The authors would like to express their sincere gratitude to Mrs. Elia Setiana from Universitas
Informatika dan Bisnis Indonesia for his valuable assistance in data preprocessing and technical validation
during the early phase of this research. The authors have obtained consent from all individuals mentioned in
this section.

FUNDING INFORMATION
This research was funded by the Research and Community Service Division (PPM) of Universitas
Informatika dan Bisnis Indonesia (UNIBI) under an internal grant of IDR 10,000,000.

XGBoost optimization using hybrid Bayesian optimization and nested cross validation for … (Budiman)
700  ISSN: 1693-6930

AUTHOR CONTRIBUTIONS STATEMENT

This journal uses the Contributor Roles Taxonomy (CRediT) to recognize individual author
contributions, reduce authorship disputes, and facilitate collaboration.

Name of Author C M So Va Fo I R D O E Vi Su P Fu
Budiman ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Nur Alamsyah ✓ ✓ ✓ ✓ ✓ ✓ ✓
Titan Parama Yoga ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
R. Yadi Rakhman ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Alamsyah
Elia Setiana ✓ ✓ ✓ ✓ ✓

C : Conceptualization I : Investigation Vi : Visualization

M : Methodology R : Resources Su : Supervision
So : Software D : Data Curation P : Project administration
Va : Validation O : Writing - Original Draft Fu : Funding acquisition
Fo : Formal analysis E : Writing - Review & Editing

CONFLICT OF INTEREST STATEMENT

Authors state no conflict of interest.

DATA AVAILABILITY
The data that support the findings of this study are openly available in Kaggle at
https://www.kaggle.com/datasets/arashnic/fitbit, reference title: Fitbit Fitness Tracker Data.

REFERENCES
[1] C. Del-Valle-Soto, R. A. Brisen˜o, L. J. Valdivia, and J. A. Nolazco-Flores, “Unveiling wearables: exploring the global
landscape of biometric applications and vital signs and behavioral impact,” BioData Mining, vol. 17, no. 1, p. 15, 2024, doi:
10.1186/s13040-024-00368-y.
[2] A. Fucarino et al., “Emerging technologies and open-source platforms for remote physical exercise: Innovations and
opportunities for healthy population—a narrative review,” in Healthcare, vol. 12, no. 15, p. 1466, 2024, doi:
10.3390/healthcare12151466.
[3] G. Ji, J. Woo, G. Lee, C. Msigwa, D. Bernard, and J. Yun, “A IoT-Based Smart Healthcare in Everyday Lives: Data Collection
and Standardization From Smartphones and Smartwatches,” IEEE Internet of Things Journal, vol. 11, no. 16, pp. 27597-27619,
2024, doi: 10.1109/JIOT.2024.3400509.
[4] K.-R. Hong, I.-W. Hwang, H.-J. Kim, S.-H. Yang, and J.-M. Lee, “Apple watch 6 vs. galaxy watch 4: A validity study of step-
count estimation in daily activities,” Sensors, vol. 24, no. 14, p. 4658, 2024, doi: 10.3390/s24144658.
[5] V. M. P. Cortes, A. Chatterjee, and D. Khovalyg, “Dynamic personalized human body energy expenditure: Prediction using time
series forecasting lstm models,” Biomedical Signal Processing and Control, vol. 87, 2024, doi: 10.1016/j.bspc.2023.105381.
[6] L.-I. Coman, M. Ianculescu, El.-A. Paraschiv, A. Alexandru, and I.-A. Bădărău, “Smart Solutions for Diet-Related Disease
Management: Connected Care, Remote Health Monitoring Systems, and Integrated Insights for Advanced Evaluation,” Applied
Sciences, vol. 14, no. 6, p. 2351, 2024, doi: 10.3390/app14062351.
[7] M. Eed, A. A. Alhussan, A. T. Qenawy, A. M. Osman, A. M. Elshewey, and R. Arnous, “Potato consumption forecasting based
on a hybrid stacked deep learning model,” Potato Research, pp. 1–25, 2024, doi: 10.1007/s11540-024-09764-7.
[8] N. Alamsyah, Budiman, T. P. Yoga, and R. Y. R. Alamsyah, “A stacking ensemble model with smote for improved imbalanced
classification on credit data,” TELKOMNIKA (Telecommunication Computing Electronics and Control), vol. 22, no. 3, pp. 657–
664, 2024, doi: 10.12928/TELKOMNIKA.v22i3.25921.
[9] I. Priyana, N. Alamsyah, Budiman, A. P. Sarifiyono, and E. Rusnendar, “Predictive Boosting for Employee Retention with
SMOTE and XGBoost Hyperparameter Tuning,” 2024 International Conference on Smart Computing, IoT and Machine
Learning (SIML), Surakarta, Indonesia, 2024, pp. 92-97, doi: 10.1109/SIML61815.2024.10578116.
[10] M. Golyadkin, A. Gambashidze, I. Nurgaliev, and I. Makarov, “Refining the ONCE Benchmark with Hyperparameter Tuning,”
IEEE Access, vol. 12, pp. 3805-3814, 2024, doi: 10.1109/ACCESS.2023.3348750.
[11] N. U. Gilal et al., “Evaluating machine learning technologies for food computing from a data set perspective,” Multimedia Tools
and Applications, vol. 83, no. 11, pp. 32041–32068, 2024, doi: 10.1007/s11042-023-16513-4.
[12] Y. -T. Hwang, Y. -R. Hsu, and B. -S. Lin, “Using B -Spline Model on Depth Camera Data to Predict Physical Activity Energy
Expenditure of Different Levels of Human Exercise,” in IEEE Transactions on Human-Machine Systems, vol. 54, no. 1, pp. 79-
88, Feb. 2024, doi: 10.1109/THMS.2023.3349030.
[13] K.-S. Lee, “Multi-spectral food classification and caloric estimation using predicted images,” Foods, vol. 13, no. 4, p. 551, 2024,
doi: 10.3390/foods13040551.
[14] H. Wu, “Intrusion Detection Model for Wireless Sensor Networks Based on FedAvg and XGBoost Algorithm,” International
Journal of Distributed Sensor Networks, no. 1, p. 5536615, 2024, doi: 10.1155/2024/5536615.

TELKOMNIKA Telecommun Comput El Control, Vol. 23, No. 3, June 2025: 694-702
TELKOMNIKA Telecommun Comput El Control  701

[15] A. G. Putrada, N. Alamsyah, S. F. Pane, and M. N. Fauzan, “XGBoost for IDS on WSN Cyber Attacks with Imbalanced Data,”
2022 International Symposium on Electronics and Smart Devices (ISESD), Bandung, Indonesia, 2022, pp. 1-7, doi:
10.1109/ISESD56103.2022.9980630.
[16] M. T. Aziz, R. Sudheesh, R. D. C. Pecho, N. U. A. Khan, A. Ull, H. Era, and M. A. Chowdhury, “Calories burnt prediction using
machine learning approach,” Current Integrative Engineering, vol. 1, no. 1, pp. 29–36, 2023, doi:
10.59762/cie570390541120231031130323.
[17] N. Alamsyah, Saparudin, and A. P. Kurniati, “A Novel Airfare Dataset To Predict Travel Agent Profits Based On Dynamic
Pricing,” 2023 11th International Conference on Information and Communication Technology (ICoICT), Melaka, Malaysia,
2023, pp. 575-581, doi: 10.1109/ICoICT58202.2023.10262694.
[18] A. Alyaseen, A. Poddar, N. Kumar, P. Sihag, D. Lee, and T. Singh, “Assessing the compressive and splitting tensile strength of
self-compacting recycled coarse aggregate concrete using machine learning and statistical techniques,” Materials Today
Communications, vol. 38, p. 107970, 2024, doi: 10.1016/j.mtcomm.2023.107970.
[19] Z. H. Wang, Y. F. Liu, T. Wang, J. G. Wang, Y. M. Liu, and Q. X. Huang, “Intelligent prediction model of mechanical
properties of ultrathin niobium strips based on xgboost ensemble learning algorithm,” Computational Materials Science, vol.
231, p. 112579, 2024, doi: 10.1016/j.commatsci.2023.112579.
[20] A. Bigdeli, A. Maghsoudi, and R. Ghezelbash, “A comparative study of the XGBoost ensemble learning and multilayer
perceptron in mineral prospectivity modeling: a case study of the Torud-Chahshirin belt, NE Iran,” Earth Science Informatics,
vol. 17, no. 1, pp. 483–499, 2024, doi: 10.1007/s12145-023-01184-4.
[21] S. B. Jabeur, S. Mefteh-Wali, and J.-L. Viviani, “Forecasting gold price with the XGBoost algorithm and SHAP interaction
values,” Annals of Operations Research, vol. 334, no. 1, pp. 679–699, 2024, doi: 10.1007/s10479-021-04187-w.
[22] M. Y. Junior, R. Z. Freire, L. O. Seman, S. F. Stefenon, V. C. Mariani, and L. dos S. Coelho, “Optimized hybrid ensemble
learning approaches applied to very short-term load forecasting,” International Journal of Electrical Power & Energy Systems,
vol. 155, p. 109579, 2024, doi: 10.1016/j.ijepes.2023.109579.
[23] D. Tarwidi, S. R. Pudjaprasetya, D. Adytia, and M. Apri, “An optimized xgboost-based machine learning method for predicting
wave run-up on a sloping beach,” MethodsX, vol. 10, p. 102119, 2023, doi: 10.1016/j.mex.2023.102119.
[24] S. Demir and E. K. Sahin, “An investigation of feature selection methods for soil liquefaction prediction based on tree-based
ensemble algorithms using AdaBoost, gradient boosting, and XGBoost,” Neural Computing and Applications, vol. 35, no. 4, pp.
3173–3190, 2023, doi: 10.1007/s00521-022-07856-4.
[25] T. O. Omotehinwa and D. O. Oyewola, “Hyperparameter optimization of ensemble models for spam email detection,” Applied
Sciences, vol. 13, no. 3, p. 1971, 2023, doi: 10.3390/app13031971.
[26] Y. Rimal and N. Sharma, “Hyperparameter optimization: a comparative machine learning model analysis for enhanced heart
disease prediction accuracy,” Multimedia Tools and Applications, vol. 83, no. 18, pp. 55091–55107, 2024, doi: 10.1007/s11042-
023-17273-x.
[27] A. M. Elshewey, M. Y. Shams, N. El-Rashidy, A. M. Elhady, S. M. Shohieb, and Z. Tarek, “Bayesian optimization with support
vector machine model for Parkinson disease clas- sification,” Sensors, vol. 23, no. 4, p. 2085, 2023, doi: 10.3390/s23042085.
[28] B. Asadi and R. Hajj, “Prediction of asphalt binder elastic recovery using tree-based ensemble bagging and boosting models,”
Construction and Building Materials, vol. 410, 2024, doi: 10.1016/j.conbuildmat.2023.134154.

BIOGRAPHIES OF AUTHORS

Budiman received his S.T. degree in Informatics Engineering from Pasundan

University, in 2006, and his M.Kom. degree in Information Systems in 2013 from STMIK
LIKMI, Indonesia. Currently, he is also a lecturer and researcher at Universitas Informatika
dan Bisnis Indonesia at the Faculty of Technology and Informatics. His current research
interests include machine learning, deep learning, data science, and system applications. He
can be contacted at email: [email protected].

Nur Alamsyah received his S.T. degree in Informatics Engineering from Sekolah
Tinggi Manajemen Informatika Bandung, in 2002, and his M.Kom. degree in Information
Systems in 2013 from Sekolah Tinggi Manajemen Informatika LIKMI Bandung, Indonesia.
He is currently continuing his Doctor of Informatics Study at Telkom University Bandung. He
is also a lecturer and researcher at Universitas Informatika dan Bisnis Indonesia at the Faculty
of Technology and Informatics. His current research interests include data science, machine
learning, deep learning, optimization and text mining. He can be contacted at email:
[email protected].

XGBoost optimization using hybrid Bayesian optimization and nested cross validation for … (Budiman)
702  ISSN: 1693-6930

Titan Parama Yoga received his S.Kom. degree in accounting computerization

(specialization in informatics management) from STMIK Bandung, in 2001 and his M.Kom.
degree in information systems in 2016 from STMIK LIKMI. Currently he is also a lecturer and
researcher at the University of Informatics and Business Indonesia at the Faculty of
Technology and Informatics. His current research interests include information systems audit,
UI/UX, and data science. He can be contacted at email: [email protected].

R. Yadi Rakhman Alamsyah received his S.T. degree in Informatics

Engineering from STMIK Bandung, in 2007, and his M.Kom. degree in Information Systems
in 2017 from STMIK LIKMI, Indonesia. Currently, he is also a lecturer and researcher at
Universitas Informatika dan Bisnis Indonesia at the Faculty of Technology and Informatics.
His current research interests include software engineering, algorithm data, and multimedia.
He can be contacted at email: [email protected].

Elia Setiana received her S.Kom. degree in Informatics from Langlangbuana

University Bandung in 2000 and her M.T. degree in the same field in 2015. Currently, she is
also a lecturer and researcher at Universitas Informatika dan Bisnis Indonesia in the Faculty of
Technology and Informatics. His current research interests include data science and software
engineering. He can be contacted at email: [email protected].

TELKOMNIKA Telecommun Comput El Control, Vol. 23, No. 3, June 2025: 694-702