Decision support system in machine learning models for a face recognition-based attendance system

TELKOMNIKA Telecommunication Computing Electronics and Control
Vol. 23, No. 2, April 2025, pp. 371~381
ISSN: 1693-6930, DOI: 10.12928/TELKOMNIKA.v23i2.26412  371
Journal homepage: http://telkomnika.uad.ac.id
Decision support system in machine learning models for a face
recognition-based attendance system
Joseph Teguh Santoso1,2
, Danny Manongga2
, Hendry2
1
Department of Computer Science, Faculty of Academic Studies, Universitas Sains dan Teknologi Komputer, Semarang, Indonesia
2
Department of Computer Science, Faculty of Information Technology, Satya Wacana Christian University, Salatiga, Indonesia
Article Info ABSTRACT
Article history:
Received Jun 18, 2024
Revised Dec 2, 2024
Accepted Dec 26, 2024
This research aims to develop a predictive model using face recognition-
based attendance data and integrating decision support system (DSS) theory
with machine learning (ML) techniques to identify high-performing teachers
at vocational high schools (SMKs). The novelty of this research lies in
integrating theory with the use of face recognition data and ML algorithms
to predict and identify high-performing teachers, thereby enhancing
decision-making processes and teacher performance management in SMK
schools. The dataset consists of SMK teachers' attendance data obtained
through a face recognition attendance system, totaling 998 entries. This
research employs sensitivity analysis concepts from DSS theory and
classification approaches from ML models utilizing support vector machine
(SVM), decision trees (DT), and random forest (RF). The models are trained
and tested on Google Colab using Python, with data distribution guided by
the Pareto principle. The research findings indicate that integrating DSS
theory with ML contributes to innovation and benefits in improving
decision-making and teacher performance management by successfully
predicting high-performing teachers. Evaluation results show the highest
accuracy rate of 98% with the RF model, making it the best predictive model
compared to the other two models.
Keywords:
Attendance system
Decision making
Decision support system
Face recognition
Machine learning
This is an open access article under the CC BY-SA license.
Corresponding Author:
Joseph Teguh Santoso
Department of Computer Science, Faculty of Information Technology, Satya Wacana Christian University
St. Diponegoro No. 52-60, 50711 Salatiga, Indonesia
Email: joseph_teguh@stekom.ac.id
1. INTRODUCTION
The use of technology in optimizing human resource management processes has become a primary
focus in various fields. Especially in the current digital era, where technology-based attendance systems have
been widely adopted, such as fingerprint, smart cards, radio frequency identification (RFID), QR code, and
face recognition. Recently, face recognition technology has become a popular alternative for monitoring
individual attendance. This is evidenced by the increasing number of studies focusing on the implementation
of face recognition across various fields and institutions. In the financial sector, Srivastava and Bag [1] used
face recognition as a modern marketing tool, Joo et al. [2] developed face recognition for payment method
confirmation, and Khaparde et al. [3] used face recognition for fraud detection. Additionally, bibliometric
analysis data using Vosviewer related to the use of face recognition shows that this topic has gained
significant attention in recent years, particularly in the area of attendance systems. This is indicated by
several emerging studies such as [4]-[7] which utilize face recognition technology for attendance in various
institutions. Face recognition is also starting to be integrated with new technologies like machine learning
(ML) for specific needs. Several recent studies [6], [8]-[13] use ML for accuracy tasks in face recognition

 ISSN: 1693-6930
TELKOMNIKA Telecommun Comput El Control, Vol. 23, No. 2, April 2025: 371-381
372
detection. Additionally, research by [14]-[17] use various machine-learning models as tools for analysis in
face recognition.
About predictive models, ML has brought significant benefits in various contexts, including
employee performance prediction, sales prediction, risk analysis, and more. Several recent studies [18]-[20]
utilized ML as a predictive model for the sales and purchase of goods, while [21] employed ML as a tool for
predicting customer purchasing habits. Additionally, research by [22]-[24] used ML to predict and analyze
risks. Furthermore, studies focus on developing more complex predictive models to predict individual
attendance by considering various factors such as weather conditions, employee schedules, and other external
factors. Research by [25], [26] combined various data with weather data using ensemble learning techniques
to predict attendance and purchases with higher accuracy. However, there has been no research utilizing
employee attendance data from attendance systems themselves, indicating that the use of attendance data is
still an area that needs to be explored further.
In the context of secondary education, having accomplished teachers and efforts to support their
development are vital aspects in ensuring the quality of education provided. Additionally, the use of
attendance data in school attendance systems to predict the performance of vocational high school (SMK)
teachers has not been studied. On the other hand, there are not many studies that integrate these analytical
results into systems that directly support decision-making. In this regard, the appropriate use of theory will
support the development of predictive models to become more holistic and efficient. Decision-making
requires a set of concepts, principles, and methods to design, develop, and implement systems, one of which
is the decision support system (DSS) theory. The main goal of this theory is to provide effective support for
decision-makers in dealing with complex and unstructured problems. Concerning ML, this theory can
support the development of predictive models to become more holistic and efficient. DSS theory itself has
been applied in various fields, for instance, in the financial sector, DSS is used to investigate components in
the fintech ecosystem [27]. In decision-making, DSS is implemented on computers for automated decision-
making [28]-[32] also utilized DSS to assist web-based decision-making. In newer technology, several
studies [33]-[35] integrated DSS theory with artificial intelligence (AI) as a tool for disease detection or
diagnosis.
Although the application of DSS theory is evident in various fields, no research has been found that
uses this theory in conjunction with ML for predictive tasks simultaneously. Additionally, the use of DSS in
the field of education has not been explored. In the context of secondary education, the use of attendance data
for strategic decision-making is still very limited. Many schools record teacher attendance using face
recognition systems, but the stored data is rarely utilized for data-driven decision-making. To fill this gap,
this study aims to develop a predictive model to identify outstanding teachers and those who need further
development using ML techniques based on attendance data obtained through a face recognition-based
attendance system, integrating concepts from DSS theory. DSS can provide a structured framework for
analyzing data and offering informative recommendations for school management. Through the integration
of the selected theory, the resulting predictive model is expected to optimize performance management and
prediction accuracy in teacher classification while providing more holistic and applicable insights in the
context of educational organizations. Thus, this research aims not only to fill a gap in the academic literature
but also to make a tangible contribution to improving the effectiveness of teacher and staff performance
management in the educational context.
2. METHOD
2.1. Approach and technique
This research employs a sensitivity analysis approach (DSS theory) and ML classification
techniques for predictive tasks. DSS theory is a collection of concepts, principles, and methods used to
design, develop, and implement systems that aid in decision-making. The primary goal of this theory is to
provide effective support for decision-makers in addressing complex and unstructured problems. DSS
encompasses several approaches with different purposes and functions, including sensitivity analysis, risk
analysis, and multi-criteria analysis [36]. In this study, the sensitivity analysis approach from DSS theory is
chosen for the development of a predictive model using ML. This approach is capable of understanding how
sensitive the ML model is to changes in input parameters, which is very helpful in identifying the most
influential variables or features in the prediction results. Additionally, sensitivity analysis in DSS theory
allows for adjustments or improvements to the model if there are variables with significant impacts on the
prediction or if weaknesses are revealed during sensitivity analysis. The ML classification technique is used
to predict outstanding SMK teachers and those who need further development based on the attendance
system dataset obtained through face recognition. Classification is a task where the system is given an input
and subsequently classifies that input into the appropriate category or class [37]. The classification approach

TELKOMNIKA Telecommun Comput El Control 
Decision support system in machine learning models for a face recognition-based … (Joseph Teguh Santoso)
373
in the ML model developed in this research uses support vector machine (SVM), decision tree (DT), and
random forest (RF) models. These three models are chosen due to their performance accuracy, ability to
handle various types of data (including categorical and numerical data), model flexibility, and scalability.
2.2. Integration of decision support system theory in the development of machine learning predictive
models
Integrating concepts and techniques from DSS theory into the development of ML models can
enhance the system’s ability to provide better decision-making support. Figure 1 shows the integration steps
performed in the ML model. In the identification phase, research needs and existing problems are identified
and understood for decision-making regarding which ML model to use. Once the problem is identified, data
characteristics, problem complexity, and analysis objectives are considered when selecting the ML model.
During the data collection step, teacher attendance data is gathered through a face recognition-based
attendance system implemented in SMKs. Subsequently, the data undergoes several processing stages to
generate relevant features through preparation, cleaning, and feature extraction as needed for model training
and testing. Following this stage, the development of the ML model proceeds with training and testing, and
predictions are made using the prepared data.
Figure 1. Integration model of DSS theory into the developed ML model
2.3. Dataset
The dataset consists of data extracted in MySQL database format, stored using Excel format, and
subsequently processed according to research needs before being saved in CSV format to support data usage
in the training and testing of ML models. The dataset extracted from the system for research purposes totals
998 entries. Additional data includes performance data of SMK teachers obtained from performance
evaluations, classroom activities, and interviews conducted with several staff members including teachers,
principals, and SMK students. The dataset used in the research undergoes preliminary analysis and
processing before use. Data cleaning involves filling in missing values to avoid skewed data distribution and
prevent overfitting. Subsequently, the data is labeled and separated according to the required features.

 ISSN: 1693-6930
374
The dataset is then divided into training and testing data using the Pareto principle or the 80/20 rule [38],
where 80% of the data is used for model training and 20% for model testing and evaluation of the developed
model’s performance. In this model development step, 80% of the dataset is trained using SVM, DT, and RF
models to enable the models to learn patterns present in the dataset. Various model parameters are adjusted to
enhance performance and prevent overfitting. Next, model testing is conducted using the remaining 20% of
the dataset to evaluate model performance, involving measurement metrics such as accuracy, precision,
recall, and F1 score.
2.4. Evaluation metrics
2.4.1. Accuracy
This is a metric for evaluating the performance of classification models, especially when the target
classes are balanced. This metric measures how many correct predictions are made out of the total
predictions made by the models (SVM, DT, and RF), providing an overall picture of how well the models
predict. The accuracy metric is calculated using the (1):
𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 =
𝑇𝑃+𝑇𝑁
𝑇𝑃+𝑇𝑁+𝐹𝑃+𝐹𝑁
(1)
where true positive (TP) is the number of correct predictions for the positive class. True negative (TN) is the
number of correct predictions for the negative class. False positive (FP) is the number of incorrect predictions
for the positive class. False negative (FN) is the number of incorrect predictions for the negative class.
2.4.2. Recall
This metric will measure the model’s ability to find all positive samples and show how well it
effectively identifies high-performing teachers. The recall metric will be calculated using (2), which ensures
that the model’s capability to identify relevant samples is quantified precisely. High recall indicates that the
model is successful in minimizing false negatives, ensuring that most high-performing teachers are correctly
identified.
𝑅𝑒𝑐𝑎𝑙𝑙 =
𝑇𝑃
𝑇𝑃+𝐹𝑁
(2)
2.4.3. Precision and F1-score
The precision metric measures how accurate the model is in predicting the positive class, i.e., how
many of the positive predictions are positive. F1-score is a harmonic mean of recall and precision that will
provide a comprehensive overview of the developed model’s performance by considering recall and
precision simultaneously. Precision metric is calculated using (3), and for the F1-score metric, it is calculated
using (4).
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 =
𝑇𝑃
𝑇𝑃+𝐹𝑃
(3)
𝐹1 − 𝑆𝑐𝑜𝑟𝑒 = 2 ×
𝑃𝑟𝑒𝑠𝑖𝑠𝑖 ×𝑅𝑒𝑐𝑎𝑙𝑙
𝑃𝑟𝑒𝑠𝑖𝑠𝑖+𝑅𝑒𝑐𝑎𝑙𝑙
(4)
2.5. Training and testing environment
The training and testing of the model are conducted using the Python programming language with
the scikit-learn library in the Google Colab environment. Python is a highly capable programming language
for data analysis in AI. Meanwhile, the scikit-learn library is powerful and easy to use for various machine-
learning algorithms. This library will be used to implement SVM, DT, and RF algorithms. Loading datasets
in comma separated value (.CSV) format in Google Colab is facilitated using Pandas, an open-source
software library used in the Python programming language for data manipulation and analysis. This library
provides data structures and data analysis tools that are easy to use, especially in the form of DataFrames.
Pandas enable users to clean, manipulate, and analyze data more efficiently.
2.6. Training and testing dataset
Figure 2 shows the training and testing dataset using a ML model. In Google Colab, the required
libraries are imported first (pandas, sklearn, SVM model, RF, DT, and also the necessary metrics). Using the
CSV format, the dataset is uploaded into Google Colab using Python. In this case, the dataset was prepared
using separated labels and features ready for pandas using “&(cell)&” in Excel in CSV format. This dataset
contains attendance data of SMK teachers for the last 3 months obtained through a face recognition-based

375
attendance system. Besides labels, the dataset also includes several features such as frequency of attendance,
late arrival date, number of working days, and others. For separating features and labels in the uploaded
dataset, the command ‘data’ becomes #Features and #Labels, X=data.drop(‘Performance’, axis=1) # Features
and y=data[‘Performance’] # Labels, where ‘Performance’ is the label in the table with a value of 1 for high-
performing SMK teachers and 0 for those needing support. Regarding the division of data for training and
testing, this research follows the Pareto principle (80/20 rule) [38].
Figure 2. Training, testing, and prediction process of the developed ML model in the research
In model training, SVM, DT, and RF models are initialized and then trained using the training data.
During model prediction, the testing data is used. If no error message appears, the model evaluation step can
proceed. However, if an error message appears in the prompt, the size and content of the array are checked
using the ‘shape’ attribute. If there is an imbalance in the data at the indices that appear, the size is adjusted
to balance it. Further in model evaluation, each model is evaluated using predefined metrics to determine
accuracy, recall, precision, and the F1-score value of each model. If no further issues arise, the evaluation
results are displayed. The ValueError that occurs during model prediction indicates a mismatch in the number
of samples in the input variables provided to the “accuracy_score()” function. When using the function
accuracy_score(y_test, svm_pred), both arguments y_test and svm_pred for the SVM model being used must
have the same number of samples. In this normalization test, several analyses are conducted including
descriptive analysis, outlier tests, and non-parametric analysis until the data is normally distributed and ready
for training, testing, and predicting results as needed. Additionally, Shapiro-Wilk and Kolmogorov-Smirnov
tests are conducted to assess the normality of the data distribution.

 ISSN: 1693-6930
376
3. RESULTS AND DISCUSSION
3.1. Machine learning model
Table 1 shows the results of evaluating 3 ML models in this study, showing an interesting
performance comparison in predicting the performance of SMK teachers who excel and those who need
development. According to the findings presented in Table 1, the SVM model achieved high accuracy,
reaching 97.2%, and perfect precision of 100%. This indicates that the model correctly classifies teachers
approximately 97.2% of the time and identifies all teachers who are excelling or in need of development. The
recall score of the SVM model is 95.7%, slightly below that of RF and DT. Meanwhile, the F1 score for
SVM is the lowest at 97.8%. On the other hand, the DT model achieved the same accuracy level as the RF, at
98.3%, with precision and recall reaching 100% and 96.4%, respectively. Although its precision and recall
are slightly lower than RF, the DT F1 score is 97.7%, indicating good overall performance.
Table 1. The results of the metric measurements on the three machine-learning models used
ML model
Measurement metrics
Accuracy Precision Recall F1-score
SVM 0.972 1.0 0.957 0.978
RF 0.983 1.0 0.974 0.987
DT 0.983 1.0 0.964 0.977
Figure 3 shows the heatmap diagram regarding the performance metrics of the models. In the RF
model, despite having equally high accuracy as SVM at 98.3%, this model shows slightly lower precision
and recall compared to SVM, with precision and recall reaching 100% and 97.4% respectively. Additionally,
the F1 score for the RF model reaches 98.7%, indicating very good overall performance. Therefore, although
all three models have high accuracy, RF stands out with higher precision and F1 score, indicating that RF is
the most effective model in identifying high-performing employees based on attendance data.
Figure 3. Heatmap diagram regarding the performance metrics of the ML models used for predicting high-
performing teachers
3.2. Sensitivity analysis of the decision support system concept
Figure 4 shows the sensitivity analysis results of this study. Sensitivity analysis of the DSS theory
was conducted using key parameters on three ML models: SVM, RF, and DT. Figure 5 shows the sensitivity
analysis for each model. For the SVM model, variations were performed on the parameter C, which controls
the decision boundary hardness. The results show that SVM accuracy tends to remain stable and quite high
across most tested values of C, with slight decreases at smaller C values. This suggests that SVM may not be
overly sensitive to the C parameter in this case. In contrast, for RF, variations were done on the number of
trees (n_estimators). It was found that RF accuracy tends to increase with higher numbers of trees, indicating
that an ensemble of many trees generally provides better performance, making RF advantageous for
classification problems. Meanwhile, for the DT, variations were conducted on the tree depth. The results
indicate that DT accuracy tends to increase with deeper trees up to a certain point, but beyond that point,
accuracy starts to decrease, suggesting overfitting. Therefore, the trade-off between increasing accuracy and

377
the risk of overfitting when determining tree depth needs to be carefully considered. Overall, these findings
highlight the sensitivity of each model to its respective parameters and emphasize the importance of
parameter tuning for optimizing model performance in practical applications.
Figure 4. The result of sensitivity analysis for ML model used for high-performing predictions using DSS
theory
Figure 5. The result of sensitivity analysis for each model

 ISSN: 1693-6930
378
The sensitivity analysis further indicates that the SVM model demonstrates the most consistent
performance, exhibiting an increase in accuracy with adjustments to the C parameter. This increase
underscores the model’s robustness, particularly in high-stakes predictions where precision is critical.
Conversely, the DT model reveals that its accuracy is optimal at certain depths, necessitating careful
parameter tuning to avoid overfitting. The RF model, characterized by its ensemble approach, shows
improved accuracy with an increased number of trees, reinforcing the notion that ensemble methods can
enhance predictive performance. These insights collectively emphasize the necessity of understanding model
sensitivities in the context of DSS theory, highlighting the significance of strategic parameter adjustments to
optimize ML outcomes in practical applications.
3.3. Normality test
In the previously used dataset consisting of 998 data points, it was found that the data was not
normally distributed, prompting further analysis. In this study, data normalization was conducted using an
outlier function to remove unnecessary data points. Subsequently, tests and analyses were performed to
ensure the success of normalization and to verify if the data met the assumptions required for further analysis.
In this case, the dataset initially containing 998 data points was reduced to 876 data points ready for use.
Descriptive analysis was then carried out to understand the characteristics of the normalized data distribution
post-outlier removal. Table 2 shows the Shapiro-Wilk and Kolmogorov-Smirnov p-value tests on the models
used. The normality of the residual distribution was assessed through regression analysis to fulfill the
assumptions required by the method used. Next, Shapiro-Wilk and Kolmogorov-Smirnov tests were
conducted on the evaluation data from each model. The Shapiro-Wilk test results showed that the p-values
for all three models (SVM, RF, and DT) were well above the significance level of 0.05, indicating that the
evaluation data from all three models tended to follow a normal distribution.
Table 2. Shapiro-Wilk and Kolmogorov-Smirnov p-value tests on the models used for prediction with the
integration of DSS theory
No Model Shapiro-Wilk p-value Kolmogorov-Smirnov p-value
1 SVM 0.890903 0.001633
2 RF 0.899373 0.001476
3 DT 0.899373 0.001476
On the other hand, the Kolmogorov-Smirnov test indicated that the p-values were significantly
below the 0.05 significance level, suggesting that the evaluation data from all three models did not
significantly differ from a normal distribution. Therefore, it can be concluded that the evaluation results
indicate very good performance from all three classification models, while the evaluation data tends to follow
a normal distribution based on the Shapiro-Wilk test, despite minor differences indicated by the
Kolmogorov-Smirnov test. This provides additional confidence in interpreting the evaluation results and the
reliability of the conclusions drawn from the model performance analysis.
3.4. Recommendation
From the research findings, SVM showed an accuracy of 97.2%, with a maximum precision of
100%, identical to the other two models, and a recall value of 95.7%. However, its F1 score was quite high,
reaching 97.8%. On the other hand, DT achieved an accuracy of 98.3% with a recall score of 96.4% and an
F1-score of 97.7%. In this context, RF exhibited the highest accuracy compared to the other two models, at
98.3%, with a recall of 97.4%. This indicates that the RF model can provide very good predictions in
identifying SMK teachers who excel and those who need additional guidance. While SVM and DT also
performed well, RF strikes an optimal balance between accuracy, precision, and recall. Therefore, it is highly
recommended to use a RF as the most suitable model for this prediction task. Moreover, this model can also
be employed in cases where maximizing accuracy and precision is crucial, while a DT may be considered if
computational speed is a primary factor and there is tolerance for slight decreases in the F1 score.
3.5. Discussion
The results of this study indicate that the RF model outperforms others in predicting the
performance of high-performing vocational school (SMK) teachers and those who require further
development. The highest accuracy achieved by this model was 98.3%, with near-perfect precision and recall
values of 100% and 97.4%, respectively. This positions RF as the most effective predictive model compared
to the SVM and DT models. Several previous studies have highlighted the advantages of RF in handling
large datasets and decision-making tasks. For instance, Khiavi [26] used RF to predict base flow generation

379
potential in flooded areas and found that RF outperformed other methods such as linear regression and DT
models in terms of accuracy. This aligns with our findings, where RF demonstrated superior reliability in
handling complex data like face recognition-based attendance data. Additionally, a study by [10], which
developed a face recognition algorithm for fraud detection systems, revealed that integrating ML with RF
provided more accurate and reliable results compared to other standalone models. These findings support our
results, where RF effectively handled variations in attendance data and produced more stable predictions.
While the SVM model showed close accuracy to RF at 97.2%, it appeared less sensitive to data
variations, particularly with a slightly lower recall of 95.7%. This is consistent with the findings of [8], who
found that SVM was less efficient in handling complex or nonlinear features such as attendance data. Their
study highlighted that although SVM performs well in binary classification tasks, the model requires precise
parameter adjustments to function optimally, especially with heterogeneous datasets. The DT also
demonstrated solid performance with an accuracy of 98.3%, but this model is more prone to overfitting,
particularly with high-dimensional data. Chavan and Sherekar [9] observed similar issues in their use of DT
for face recognition ML tasks, where the model performance degraded when faced with highly variable data.
In our study, the DT achieved the same precision as the RF, but with a slightly lower recall of 96.4%,
indicating that the model was less effective at identifying all teachers requiring development.
Integrating DSS theory in developing ML models has significantly enhanced prediction accuracy
and decision-making capabilities. DSS helps identify the most influential input parameters, allowing dynamic
model adjustments based on input data variations. A study by [32] demonstrated that integrating DSS in web-
based systems improved decision-making accuracy by offering better sensitivity analysis to changing inputs.
Our research corroborates these findings, showing that DSS improves prediction performance by guiding
model parameter selection, particularly for critical variables such as teacher attendance and lateness.
The findings of this research have significant implications for school management, particularly in
improving teacher performance in vocational schools. The integration of face recognition-based attendance
systems with ML models and DSS can assist schools in identifying high-performing teachers and providing
recommendations for further development for those needing support. This aligns with research by [1], which
identified that using face recognition technology in education can enhance human resource management
efficiency, particularly in monitoring and evaluating staff performance. Our study emphasizes the importance
of utilizing existing attendance data for more strategic decision-making, an area that has not yet been widely
adopted in the Indonesian education system.
While this study provides valuable insights, several limitations should be noted. The dataset is
limited to SMK teachers’ attendance data over the last three months. Future research should incorporate a
larger and more diverse dataset, considering other factors such as performance evaluations from students and
principals. Additionally, this study used three ML algorithms, exploring other models such as neural
networks or ensemble learning could offer deeper insights into model reliability in different scenarios.
4. CONCLUSION
In this study, sensitivity analysis of the DSS theory was applied to ML prediction models using a
classification approach to analyze the performance and predict SMK teachers who excel and those who need
additional support. The three ML models used in this study were SVM, DT, and RF. The results showed that
all models achieved a high precision level of 100%, indicating that the models rarely make mistakes in
classifying teachers who excel and those who need additional support. However, in terms of accuracy, RF
and DT performed better compared to SVM. The accuracy rates for RF and DT were both 98.3%, while
SVM was slightly lower at 97.2%. This suggests that both RF and DT have advantages in predicting teacher
classifications. On the other hand, all three models also showed balanced results between recall and
precision, as indicated by their F1 scores. In this case, the F1 scores for all models were above 97%,
indicating that all tested models were able to achieve a good balance in identifying teachers who need
additional support (recall) and minimizing errors in classifying teachers who excel (precision). Therefore, for
the prediction task in this study, RF and DT can be considered excellent choices over SVM.
ACKNOWLEDGEMENTS
The author extends appreciation to colleagues and advisers for their valuable insights and expertise,
which were crucial to the success of this study.
REFERENCES
[1] G. Srivastava and S. Bag, “Modern-day marketing concepts based on face recognition and neuro-marketing: a review and future
research directions,” Benchmarking, vol. 31, no. 2, pp. 410–438, Feb. 2024, doi: 10.1108/BIJ-09-2022-0588.

 ISSN: 1693-6930
380
[2] K. Joo, J. Kim, and J. Hwang, “Effects of foodservice consumers’ perceptions of face recognition payment on attitude, desire, and
behavioral intentions: a cross-cultural study,” Journal of Travel & Tourism Marketing, vol. 41, no. 3, pp. 359–376, Mar. 2024,
doi: 10.1080/10548408.2024.2318429.
[3] R. P. K, A. Khaparde, V. Bendre, and J. Katti, “Fraud detection and prevention by face recognition with and without mask for
banking application,” Multimedia Tools and Applications, pp. 1–24, Apr. 2024, doi: 10.1007/s11042-024-19021-1.
[4] M. Barhate et al., “Innovative attendance tracking: facial recognition,” in 2024 International Conference on Emerging Smart
Computing and Informatics, ESCI 2024, IEEE, Mar. 2024, pp. 1–6. doi: 10.1109/ESCI59607.2024.10497364.
[5] R. Misra and S. K. Shakya, “Face recognition attendance system, smart learning, college enquiry using AI Chat-Bot,” in
International Conference on Recent Trends in Engineering & Technology (ICRTET-2023), 2023, pp. 164–170.
[6] D. V. Nagagopiraju, D. M. Deepak, M. V. Vinay, E. A. Kumar, and K. Supriya, “Machine learning and blockchain-based real-
time facial recognition attendance system,” Turkish Journal of Computer and Mathematics Education (TURCOMAT), vol. 15, no.
1, pp. 133–137, Mar. 2024, doi: 10.61841/turcomat.v15i1.14554.
[7] Y. Zhang, W. Xie, and X. Yu, “Design and implementation of liveness detection system based on improved shufflenet V2,”
Signal, Image and Video Processing, vol. 17, no. 6, pp. 3035–3043, Sep. 2023, doi: 10.1007/s11760-023-02524-z.
[8] I. Branescu, R. I. Ciobanu, C. Dobre, and C. Mavromoustakis, “Decentralized machine learning for face recognition,” in
Proceedings - 2023 22nd International Symposium on Parallel and Distributed Computing, ISPDC 2023, IEEE, Jul. 2023, pp. 1–
8. doi: 10.1109/ISPDC59212.2023.00010.
[9] S. R. Chavan and S. Sherekar, “Face recognition system using CNN architecture & its model with its detection technique using
machine learning,” in Heterogenous Computational Intelligence in Internet of Things, Boca Raton: CRC Press, 2023, pp. 209–
224. doi: 10.1201/9781003363606-14.
[10] J. K. Essel, J. A. Mensah, E. Ocran, and L. Asiedu, “On the search for efficient face recognition algorithm subject to multiple
environmental constraints,” Heliyon, vol. 10, no. 7, pp. 1–14, Apr. 2024, doi: 10.1016/j.heliyon.2024.e28568.
[11] A. Goel, A. K. Goel, and A. Kumar, “The role of artificial neural network and machine learning in utilizing spatial information,”
Spatial Information Research, vol. 31, no. 3, pp. 275–285, Jun. 2023, doi: 10.1007/s41324-022-00494-x.
[12] M. N. Naik, A. Kaur, N. Yathiraju, S. Das, and K. Pant, “Improved and accurate face mask detection using machine learning in
the crowded places,” in 2023 3rd International Conference on Advance Computing and Innovative Technologies in Engineering,
ICACITE 2023, IEEE, May 2023, pp. 572–576. doi: 10.1109/ICACITE57410.2023.10182567.
[13] S. Saleem, J. Shiney, B. P. Shan, and V. K. Mishra, “Face recognition using facial features,” Materials Today: Proceedings, vol.
80, pp. 3857–3862, 2023, doi: 10.1016/j.matpr.2021.07.402.
[14] A. Mughaid et al., “A novel machine learning and face recognition technique for fake accounts detection system on cyber social
networks,” Multimedia Tools and Applications, vol. 82, no. 17, pp. 26353–26378, Jul. 2023, doi: 10.1007/s11042-023-14347-8.
[15] Rangayya, Virupakshappa, and N. Patil, “Improved face recognition method using SVM-MRF with KTBD based KCM
segmentation approach,” International Journal of System Assurance Engineering and Management, vol. 15, no. 1, pp. 1–12, Jan.
2024, doi: 10.1007/s13198-021-01483-3.
[16] G. Sanil, K. Prakash, S. Prabhu, V. C. Nayak, and S. Sengupta, “2D-3D facial image analysis for identification of facial features
using machine learning algorithms with hyper-parameter optimization for forensics applications,” IEEE Access, vol. 11, pp.
82521–82538, 2023, doi: 10.1109/ACCESS.2023.3298443.
[17] X. Zhou and T. C. Zhu, “Survey of research on face recognition methods based on depth learning,” Journal of Physics:
Conference Series, vol. 2717, no. 1, pp. 1–7, Mar. 2024, doi: 10.1088/1742-6596/2717/1/012027.
[18] C. D. Kothapalli, G. Navya, U. Jaladhi, S. R. Sulthana, D. L. S. Kumar, and S. P. Praveen, “Predicting buy and sell signals for
stocks using bollinger bands and MACD with the help of machine learning,” in 2023 International Conference on Sustainable
Computing and Smart Systems (ICSCSS), IEEE, Jun. 2023, pp. 333–340. doi: 10.1109/ICSCSS57650.2023.10169500.
[19] Z. Y. Chen, Z. P. Fan, and M. Sun, “Machine learning methods for data-driven demand estimation and assortment planning
considering cross-selling and substitutions,” INFORMS Journal on Computing, vol. 35, no. 1, pp. 158–177, Jan. 2023, doi:
10.1287/ijoc.2022.1251.
[20] S. Lahmiri, S. Bekiros, and C. Avdoulas, “A comparative assessment of machine learning methods for predicting housing prices
using Bayesian optimization,” Decision Analytics Journal, vol. 6, pp. 1–8, Mar. 2023, doi: 10.1016/j.dajour.2023.100166.
[21] G. Chaubey, P. R. Gavhane, D. Bisen, and S. K. Arjaria, “Customer purchasing behavior prediction using machine learning
classification techniques,” Journal of Ambient Intelligence and Humanized Computing, vol. 14, no. 12, pp. 16133–16157, Dec.
2023, doi: 10.1007/s12652-022-03837-6.
[22] H. Ishibashi, “Framework for risk assessment of economic loss from structures damaged by rainfall-induced landslides using
machine learning,” Georisk: Assessment and Management of Risk for Engineered Systems and Geohazards, vol. 18, no. 1, pp.
228–243, Jan. 2024, doi: 10.1080/17499518.2023.2288606.
[23] I. Shinohara et al., “Re-tear after arthroscopic rotator cuff tear surgery: risk analysis using machine learning,” Journal of Shoulder
and Elbow Surgery, vol. 33, no. 4, pp. 815–822, Apr. 2024, doi: 10.1016/j.jse.2023.07.017.
[24] B. Zhao et al., “Prediction heavy metals accumulation risk in rice using machine learning and mapping pollution risk,” Journal of
Hazardous Materials, vol. 448, p. 130879, Apr. 2023, doi: 10.1016/j.jhazmat.2023.130879.
[25] R. Arboretti et al., “Predictive stadium attendance using machine leaning: a case study in italian football,” Journal of Machine
Intelligence and Data Science, vol. 5, pp. 1–8, 2024, doi: 10.11159/jmids.2024.004.
[26] A. N. Khiavi, “Machine learning modeling of base flow generation potential: a case study of the combined application of BWM
and Fallback bargaining algorithm,” Journal of Hydrology, vol. 636, p. 131220, Jun. 2024, doi: 10.1016/j.jhydrol.2024.131220.
[27] R. Ai, Y. Zheng, S. Yüksel, and H. Dinçer, “Investigating the components of fintech ecosystem for distributed energy investments
with an integrated quantum spherical decision support system,” Financial Innovation, vol. 9, no. 1, pp. 1–28, 2023, doi:
10.1186/s40854-022-00442-6.
[28] S. Kim, E. H. Kim, and H. S. Kim, “Physician knowledge base: clinical decision support systems,” Yonsei Medical Journal, vol.
63, no. 1, pp. 8–15, 2022, doi: 10.3349/YMJ.2022.63.1.8.
[29] T. Donovan, B. Abell, M. Fernando, S. M. McPhail, and H. E. Carter, “Implementation costs of hospital-based computerised
decision support systems: a systematic review,” Implementation Science, vol. 18, no. 1, pp. 1–18, 2023, doi: 10.1186/s13012-023-
01261-8.
[30] X. Liu et al., “Discrepancy between perceptions and acceptance of clinical decision support Systems: implementation of artificial
intelligence for vancomycin dosing,” BMC Medical Informatics and Decision Making, vol. 23, no. 1, pp. 1–9, 2023, doi:
10.1186/s12911-023-02254-9.
[31] B. Abell et al., “Identifying barriers and facilitators to successful implementation of computerized clinical decision support
systems in hospitals: a NASSS framework-informed scoping review,” Implementation Science, vol. 18, no. 1, pp. 1–20, 2023, doi:

381
10.1186/s13012-023-01287-y.
[32] H. Ryu et al., “A web-based decision support system (DSS) for hydrogen refueling station location and supply chain
optimization,” International Journal of Hydrogen Energy, vol. 48, no. 93, pp. 36223–36239, Dec. 2023, doi:
10.1016/j.ijhydene.2023.06.064.
[33] N. Khan et al., “Adoption and utilization of medical decision support systems in the diagnosis of febrile diseases: a systematic
literature review,” Expert Systems with Applications, vol. 220, p. 119638, Jun. 2023, doi: 10.1016/j.eswa.2023.119638.
[34] M. Mustafa and M. O. F. Malik, “Factors hindering solar photovoltaic system implementation in buildings and infrastructure
projects: analysis through a multiple linear regression model and rule-based decision support system,” Buildings, vol. 13, no. 7,
pp. 1–19, Jul. 2023, doi: 10.3390/buildings13071786.
[35] S. Zhang, “Consumer attitudes towards AI-based financial advice: insights for decision support systems (DSS) and technology
integration,” Journal of Internet Services and Information Security, vol. 14, no. 4, pp. 1–20, Nov. 2024, doi:
10.58346/JISIS.2024.I4.001.
[36] F. Perosa, L. F. Seitz, A. Zingraff-Hamed, and M. Disse, “Flood risk management along German rivers – a review of multi-
criteria analysis methods and decision-support systems,” Environmental Science and Policy, vol. 135, pp. 191–206, Sep. 2022,
doi: 10.1016/j.envsci.2022.05.004.
[37] P. C. Sen, M. Hajra, and M. Ghosh, “Supervised classification algorithms in machine learning: a survey and review,” in Advances
in Intelligent Systems and Computing, vol. 937, pp. 99–111, 2020, doi: 10.1007/978-981-13-7403-6_11.
[38] R. Sanders, “The pareto principle: its use and abuse,” Journal of Services Marketing, vol. 1, no. 2, pp. 37–40, Feb. 1987, doi:
10.1108/eb024706.
BIOGRAPHIES OF AUTHORS
Joseph Teguh Santoso received the M.Kom. degree in Informatics Engineering
from the Universitas Dian Nuswantoro, Semarang, Indonesia, in 2004, and the doctorate (Dr.)
in Theology from STTI, Jakarta, Indonesia, in 2019. In addition, he is a Ph.D. candidate in
Computer Science at the Faculty of Information Technology at Satya Wacana Christian
University (SWCU). He is currently a Chancellor of the University of Science and Computer
Technology (STEKOM University), Semarang, Indonesia. He has published over 30 journal
papers, 44 authored books, and 2 papers in conference proceedings. His current research
interests include face recognition, machine learning, data security, artificial intelligence,
software programming, e-commerce, software engineering, data analytics, IT project
management, and robotics. He can be contacted at email: joseph_teguh@stekom.ac.id.
Danny Manongga received the M.Sc. degree in Information Technology from
Queen Mary University, London, United Kingdom, in 1989, and the Ph.D. degree in
Management Information Systems and Operation Research from the University of Eart
Anglia, Norwich, United Kingdom, in 1996. He is a Professor of Computer Science in the
Graduate Division, Satya Wacana Christian University (SWCU), Salatiga, Indonesia. In
addition, he is serving as Dean of the Faculty of Information Technology at Satya Wacana
Christian University. He is also a professor who teaches as a lecturer at the Ph.D. level in
computer science at SWCU Salatiga. He has published over 40 journal papers. His research
interests are in information systems, knowledge management, machine learning, smart
systems, and artificial intelligence. He can be contacted at email:
danny.manongga@uksw.edu.
Hendry received the M.Kom. degree in Information Technology from the
Sepuluh Nopember Institute of Technology, Surabaya, Indonesia, in 2009, and the Ph.D.
degree in Information Management from the Chaoyang University of Technology, Taichung,
Taiwan, in 2018. In addition, he is serving as Deputy Dean of the Information Technology
Faculty at Satya Wacana Christian University (SWCU), Salatiga, Indonesia. He is also a
lecturer at the Ph.D. level in computer science at SWCU Salatiga. His research interests are in
machine learning, artificial intelligence, data mining, mobile application, deep learning,
operation systems, sentiment analysis, software engineering, and virtual reality. He can be
contacted at email: hendry@uksw.edu.

Decision support system in machine learning models for a face recognition-based attendance system

More Related Content

Similar to Decision support system in machine learning models for a face recognition-based attendance system (20)

More from TELKOMNIKA JOURNAL (20)

Recently uploaded (20)

Decision support system in machine learning models for a face recognition-based attendance system