Page 1 of 12 - Cover Page Submission ID trn:oid:::1:2903219264

Library Malang
2438 - Sidharta
Quick Submit

Quick Submit

LKC

Document Details

Submission ID

trn:oid:::1:2903219264 10 Pages

Submission Date 4,088 Words

Apr 29, 2024, 9:07 AM GMT+7

24,263 Characters

Download Date

Apr 29, 2024, 11:47 AM GMT+7

File Name

ICOBAR_A_Systematic_Literature_Review_Fingerp_Sidharta.pdf

File Size

292.6 KB

Page 1 of 12 - Cover Page Submission ID trn:oid:::1:2903219264

 Page 2 of 12 - AI Writing Overview Submission ID trn:oid:::1:2903219264

How much of this submission has been generated by AI?

*3%
Caution: Percentage may not indicate academic misconduct. Review required.

It is essential to understand the limitations of AI detection before making decisions

about a student's work. We encourage you to learn more about Turnitin's AI detection
capabilities before using the tool.
of qualifying text in this submission has been determined to be
generated by AI.

* Low scores have a higher likelihood of false positives.

Frequently Asked Questions

What does the percentage mean?

The percentage shown in the AI writing detection indicator and in the AI writing report is the amount of qualifying text within the
submission that Turnitin's AI writing detection model determines was generated by AI.

Our testing has found that there is a higher incidence of false positives when the percentage is less than 20. In order to reduce the
likelihood of misinterpretation, the AI indicator will display an asterisk for percentages less than 20 to call attention to the fact that
the score is less reliable.

However, the final decision on whether any misconduct has occurred rests with the reviewer/instructor. They should use the
percentage as a means to start a formative conversation with their student and/or use it to examine the submitted assignment in
greater detail according to their school's policies.

How does Turnitin's indicator address false positives?

Our model only processes qualifying text in the form of long-form writing. Long-form writing means individual sentences contained in paragraphs that make up a
longer piece of written work, such as an essay, a dissertation, or an article, etc. Qualifying text that has been determined to be AI-generated will be highlighted blue
on the submission text.

Non-qualifying text, such as bullet points, annotated bibliographies, etc., will not be processed and can create disparity between the submission highlights and the
percentage shown.

What does 'qualifying text' mean?

Sometimes false positives (incorrectly flagging human-written text as AI-generated), can include lists without a lot of structural variation, text that literally repeats
itself, or text that has been paraphrased without developing new ideas. If our indicator shows a higher amount of AI writing in such text, we advise you to take that
into consideration when looking at the percentage indicated.

In a longer document with a mix of authentic writing and AI generated text, it can be difficult to exactly determine where the AI writing begins and original writing
ends, but our model should give you a reliable guide to start conversations with the submitting student.

Disclaimer
Our AI writing assessment is designed to help educators identify text that might be prepared by a generative AI tool. Our AI writing assessment may not always be accurate (it may misidentify
both human and AI-generated text) so it should not be used as the sole basis for adverse actions against a student. It takes further scrutiny and human judgment in conjunction with an
organization's application of its specific academic policies to determine whether any academic misconduct has occurred.

Page 2 of 12 - AI Writing Overview Submission ID trn:oid:::1:2903219264

 Page 3 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264

Revolutionizing Biometric Security: A

Systematic Literature Review on Deep Learning
in Fingerprint Recognition for Resilient
Infrastructure and Global Innovation (SDG 9)
Mochammad Alfito Dwi Cahyonoa, Richard Wijaya Ha, Sidharta Sidhartaa *
a
Computer Science Department, School of Computer Science, Bina Nusantara
University, Jakarta 11480 Indonesia
*Corresponding Author: [email protected]

Abstract. The illustrated study delves further into the topic of deep learning
in fingerprint recognition, focusing on writing dated between 2019 and
2024. Key examples and disclosures were discovered through rigorous steps
of inspection, approval, examination, extraction, and blending, revealing
insights into the feasibility and advancements in this sector. With 22 studies
demonstrating its proficiency in tackling the complexities of fingerprint
authentication tasks, Convolutional Neural Networks (CNN) emerged as a
significant focal point. Particularly, CNNs' capacity to process and analyze
image data with exceptional accuracy, ensuring robust performance even in
challenging conditions, demonstrated their adaptability and effectiveness.
LSTM-RNN and Support Vector Machines (SVM) also showed a lot of
utility, highlighting different ways to deal with authentication issues.
Notwithstanding periodic assessment challenges, the chose articles
exhibited importance by laying out clear goals, utilizing sound exploration
philosophies, and yielding exhaustive discoveries. Information blend
uncovered obvious proof of the viability of profound learning approaches in
unique mark confirmation, especially CNN-based models, with great
execution measurements including critical AUC values going from 83% to
86.6%. Extraordinarily, CNN outflanked elective strategies as far as
exactness and mistake rates while managing complex models like
electromyogram (EMG) signals. Since they vow to improve precision,
proficiency, and security across a great many applications and spaces, these
discoveries by and large require the far reaching execution of cutting edge
learning methods in biometric validation frameworks not long from now. By
encouraging innovation in biometric authentication technologies, this
research contributes to the achievement of Sustainable Development Goal 9
(SDG 9) objectives for resilient infrastructure and inclusive societies.

Keywords— Fingerprint Recognition, Deep Learning, Convolutional

Neural Networks (CNN), Biometric Authentication, SDG 9 (Resilient
Infrastructure)

Page 3 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264

 Page 4 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264

1. INTRODUCTION
Our regular routines has expanded the requirement for solid and secure individual
verification components. Fingerprints, with their uniqueness and perpetual quality, have
arisen as a trailblazer in biometric confirmation. Biometric acknowledgment is the
methodology that involves the greater part of the recognizable proof elements [1]. Biometrics
can be either a physiological quality, for example, iris [2, 3], face [4], palm [5, 6], ear [7],
finger surface, impression and unique finger impression, or a social characteristic such as
signature, penmanship, step and voice. Profound learning calculations have the remarkable
capacity to extricate complex examples from a lot of informasi. These qualities make them
especially fit to the errand of unique mark acknowledgment, where little things-the trademark
edges and forks in fingerprints-hold the way to ID. By utilizing profound learning
engineering, finger impression acknowledgment frameworks can accomplish better
execution thought about than customary strategies. This paper digs into the entrancing
universe of unique finger impression acknowledgment utilizing profound learning. We start
with a fundamental outline of unique mark acknowledgment innovation, framing the
essential standards and constraints of traditional methodologies. Then, we leave on an
excursion into the brilliant universe of profound picking, making sense of its central ideas
and explicit applications in unique mark acknowledgment.. A significant part of this paper is
a top to bottom assessment of the benefits acquired by applying profound learning in finger
impression acknowledgment. We will cautiously take apart the enhancements in exactness,
commotion heartiness, and capacity to deal with complex unique finger impression varieties.
This conversation will be supplemented by a basic examination of the intrinsic difficulties
and potential impediments related with profound learning-based unique finger impression
acknowledgment frameworks. To give an all encompassing point of view, we will cautiously
survey existing examinations in this zone. We will feature the imaginative advances made
lately, featuring the state of the art profound learning procedures that are driving the field
forward. Moreover, we will investigate the future, investigating promising examination
bearings that can possibly additionally reform unique mark acknowledgment utilizing
profound learning.

2. METHODOLOGY

Fig. 1. Diagram of the Kitchenham & Cochrane approach

This study aims to conduct a systematic literature review (SLR). Finding, assessing, and
synthesizing literature reviews that are appropriate for study questions or themes is done
Page 4 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264
 Page 5 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264

using the systematic literature review (SLR) research approach. This SLR method's primary
goal is to locate analytical evidence in the body of current literature for use in future research,
policy creation, and decision-making. Steps that are related to the Kitchenham and Cochrane
narrative synthesis techniques are employed by our group[8]. The procedures are shown in
Figure 1.

2.1 Research Question

1) RQ1: What are the predominant deep learning algorithms commonly utilized across
the compiled articles for fingerprint recognition?
2) RQ2: To what extent do deep learning methodologies exhibit efficacy in the realm
of fingerprint recognition relative to conventional approaches, as indicated by the
literature?
3) RQ3: How do variations in dataset characteristics impact the performance of deep
learning models for fingerprint recognition?

2.2 Search Process

We glanced through an informational collection that usages cutting-edge electronic

investigation requests — the database we utilized was from Scopus — to find the fitting
pursuit data. We confined our journal search to disseminations conveyed some place in the
scope of 2019 and 2024 because significant learning has transformed into a critical area of
assessment in exceptional finger impression affirmation beginning around 2019. To
guarantee the kind of our sources, we just included scholarly papers and those that have been
peer-minded. We recalled straightforwardly accessible coherent articles and meeting
techniques for our chase. To find simply proper papers for this effective composing overview,
we played out the request using data from titles and adjusted works. Besides, to simplify it
to find huge articles that fit the limits of our survey, we furthermore used assigned keywords
like "Fingerprint Recognition Using Deep Learning."

2.3 Study Selection

The papers we have retrieved are eliminated throughout the Study Selection process.
Articles that are identical or duplicates are removed, as well as inappropriate articles. The
inclusion criteria will include publications that correspond to the title of our systematic
literature review, while the exclusion criteria will include papers that do not correspond to
the title of our systematic literature review.
1. The Inclusion Criteria
- The article is composed in the English language.
- The article is available for open access
- The article appears in conference proceedings or a scientific journal
- The article is a whole paper
- The article was released between 2019 and 2024.
2. The Exclusion Criteria
- The article does not satisfy the search criteria.
- The paper neglects to explain the deep learning architecture utilized.
- Articles in the systematic literature review (SLR)
- Retraction of the article

2.4 Quality Assessment

Page 5 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264

 Page 6 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264

The chosen articles will be inspected during the Quality Examination stage preceding
being summarized to isolate the significant data. This quality evaluation will probably survey
and find the article's level of significant worth. The quality assessment's actions were changed
from Zuiderwijk et al. [8] and laid out on thoughts made by Kitchenham and Agreements.
Among these requirements are:
- The entire research method is explained.
- A specified dataset and architecture are included in the study methodology.
- Well-defined goals for the research.
- The results of the investigation were clear.
- The findings of the research are stated in detail.

2.5 Data Extraction

The genuine summed-up information will be utilized to address the appraisal demands in
the Information Extraction Stage. Various pieces of data are expeditiously cleared out,
uniting the going with the methodology, feature extraction, dataset, year of stream, creator
name, and title. High precision is all around ordinary for extra information to take out the
huge data, like the strategy's deficiency, the assessment's next execution, and the openings of
the driven study. The separated data is then used for information blend.

2.6 Data Synthesis

The collected data will be used in the Data Mixture Stage to wrap up the survey procedure.
We use story blend, which is changed from the Cochrane technique [9], for data mix. We
eliminated critical data that was presented as a summary. To work with our response to the
investigation questions, we have composed the summary outcomes into a Succeed table.
Likewise, the aftereffects of the pursuit procedure, focus on assurance, quality appraisal, and
data extraction will be utilized to address the assessment questions.

3. RESULT & DISCUSSION

3.1 Search Process Results

Table 1. Keywords Results

Database Keywords No. Description
Paper
Scopus (“fingerprint recognition” OR 103 Focused on journals or
“fingerprint identification”) AND conference proceedings.
(“deep learning” OR “neural Filtered based on the set year
network”) AND (algorithm OR from 2019 to 2024 which has
method OR technique) an Open Access Type

Because of time imperatives, we led our hunt exclusively inside one information base,
choosing Scopus for its thorough pursuit functionalities. Our pursuit interaction involved the
usage of explicit watchwords connected with our examination subject, trailed by refinement
to meet foreordained rules. These measures incorporated the necessity for articles to be open-
access, distributed somewhere in the range of 2019 and 2024, and obtained from scientific
journal or conference proceedings. The results of this search try have been arranged and are
introduced in Table 1 for reference.

Page 6 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264

 Page 7 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264

3.2 Study Selection Results

We select given the consideration and rejection rules illustrated in segment 2.3 (Study
Selection) for the review choice discoveries. Segment 3.1 (Search Process Results) has
applied incorporation and rejection standards. While utilizing the Scopus search devices, the
accompanying measures were applied: incorporation rules no. 2, 3, and 5 (The article is
available for open access, appears in conference proceedings or a scientific journal, was
released between 2019 and 2024.). Thus, the list items in Table 1 straightforwardly satisfied
the consideration rules no. 2, 3, and 5.

We will confine our discussion to the thought and dismissal models that were not used in
Section 3.1. Several of these essentials are the thought principles numbered 1 and 4 (the
article is written in English and is a completed paper) and every preclusion condition kept in
Section 2.3. 48 articles can be used considering the way that 58 articles failed to meet the
models for dismissal principles no. 1 (The article does not satisfy the search criteria), no. 2
(The paper neglects to explain the deep learning architecture utilized), and no. 3 (Articles in
the systematic literature review (SLR)). Besides, we found a couple of articles that failed to
meet the requirements of dismissal guideline no. 4 (Article is Retracted), with an amount of
11 articles, really proposing that there are 37 articles that we can use. Table 2 shows the
revelations from the survey assurance process.

Table 2. Study Selection Results

No Inclusion criteria # of articles included
1 Written in English 437
2 Open access 109
3 Journal and conference proceedings 106
4 Full paper 106
5 Published in 2019-2024 106
No. Exclusion criteria # of articles included
1 Not match with topic 48
2 Does not explain deep learning 48
3 Systematic literature review 48
4 Retracted 37

3.3 Quality Assessment Results

After applying the specified inclusion and exclusion criteria stated in Section 2.4 (Quality
Assessments), we thoroughly analysed each selected article for relevance and
appropriateness in our thorough Systematic Writing Review. Our review sought to ensure
that severe standards were followed, notably in terms of methodological rigour, dataset
clarity, and presentation of study findings. Interestingly, none of the examined articles had
evident quality flaws. However, we meticulously found methodological errors in several
research and conducted in-depth examinations of the deep learning frameworks employed.
Despite occasional obstacles, each evaluated article demonstrated exceptional quality
standards, necessitating a more in-depth analysis within the scope of our assessment. This
careful quality control technique strengthens the credibility of our Systematic Writing
Review and emphasises our dedication to maintaining scholarly integrity when synthesising
data from the contemporary literature on deep learning in fingerprint recognition.

3.4 Data Extraction Results

Page 7 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264

 Page 8 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264

Table 3. Example Data Extraction from Paper

Title Authors Year Methodology Dataset Feature Extraction
A Hybrid Vaishali 2023 The methodology CMBD and NITG Extracting handcrafted
Model by Kamble, involves combining fingerprint databases features using DCT
Combining Manisha DCT features, of children. The coefficients' mid and
Discrete Dale, machine learning recognition rate for high frequency bands,
Cosine Vinayak classifiers, and Deep DCT features and a as well as texture
Transform Bairagi learning algorithms simple distance features (standard
and Deep for children classifier was 63% for deviations) from
Learning for fingerprint 100 subjects in an 80- fingerprint images
Children identification, 20 train-test split. The using DCT and
Fingerprint extracting NITG database CurveDCT applied to
Identification handcrafted features showed better results ROIs, resulting in a
[10] using DCT, and with DCT features and total of 36 features
validating the model a machine learning from four parts stored
on publicly available approach. as a training dataset.
databases.

The "Data Extraction Results" detail an data extraction technique, using Segment 2.5
measures. Table 3 arranges data from appropriations for overview assessment. The
examination is straightforward, thorough, and expects to give an unmistakable
comprehension of finger impression ID.

3.5 Data Synthesis Results

1) RQ1: What are the predominant deep learning algorithms commonly utilized across
the compiled articles for fingerprint recognition?

Table 4. Algorithms Classification Results

Algorithms Number of Articles
CNN 22
DNN 2
LSTM-RNN, or RNN 9
ANN 2
YOLO 1
SVM 5
VGG 3
KNN 3
MFFO 1
MLP 1
ResNet 5
EEG 1

In examining the organized articles focused on one-of-a-kind imprint affirmation,

Convolutional Neural Networks (CNN) emerge as the predominant significant learning
computations utilized, with 22 articles using this system. This marvelous inescapability
features the crucial work CNN plays in keeping an eye on the natural complexities of
extraordinary finger impression affirmation tasks. CNN's solidarity starts from its capacity
to shock to truly process and analyze picture data, a middle essential in remarkable imprint
affirmation where baffling models and fine nuances ought to be definitively gotten and
Page 8 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264
 Page 9 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264

separated. The convolutional layers inside CNN structures prevail at feature extraction from
complex finger impression pictures, enabling precise affirmation even inside seeing upheaval
or assortments. Additionally, CNN's adaptability to a variety of extreme scratching
verification scenarios refreshes its appeal and contributes to the favored choice throughout
evaluations. With nine publications arranging this approach, LSTM-RNN (or just RNN)
emerges as the second most popular assessment after CNN. Because LSTM-RNN is overly
long, it is accommodating in locating brief situations among spectacular carving information,
which contributes to its sufficiency in clear authentication efforts. Moreover, five studies
make use of this evaluation to trace down the epic use of Support Vector Machines (SVM).
SVM is a reasonable option for staggering cutting because of its ability to manipulate highly-
layered information and create a ruckus. No matter what the presence of various
computations, for instance, VGG, K-Nearest Neighbors (KNN), and ResNet, clearly CNN's
solidarity stays unmatched in the area of extraordinary finger impression affirmation
research. From Table 4, clearly couple of articles unequivocally take on or use CNN plans,
or adjusted frames thereof, featuring its adaptability and ampleness in watching out for the
uncommon troubles introduced by finger impression affirmation endeavors. One of the
articles that uses the CNN plan (BİLGİN et al., " Biometric Personal Classification with Deep
Learning Using EMG Signals " 2023) [11]. In this paper, it is represented that the significant
learning-based individual affirmation model using Electromyogram (EMG) signals achieved
basic precision levels. The Convolutional Neural Networks (CNN) system yielded a
precision of 95.57%, while the Long Short-Term Memory (LSTM) strategy achieved an
accuracy of 93.88%. In the presented table, it might be seen that CNN outmaneuvers LSTM
in describing EMG data. Besides, it is seen that the use of the Rectified Linear Unit (ReLU)
authorization capacity in the CNN configuration added to the show improvement of the
model. CNN + ReLU yielded a precision of 95.57% and an endorsement accuracy of 72.07%,
while for goofs in the CNN designing, CNN made an error of 49.22% and an endorsement
slip-up of 85.09%. On the other hand, MLP + ReLU achieved an accuracy of 26.32% and an
endorsement precision of 26.88%, while for bumbles in the MLP designing, MLP made a
mix-up of 252.73% and an endorsement error of 247.43%. For LSTM-RNN + ReLU, the
precision came to 75.13%, with an endorsement accuracy of 63.53%, while for botches in
the LSTM-RNN plan, LSTM-RNN made a screw-up of 78.68%, and the endorsement botch
was 107.49%. These data show that CNN is all the more impressive in portraying EMG
signals diverged from LSTM. This asserts that the CNN design is more proper for managing
complex models in EMG signals, and the results frame the capacity of the CNN configuration
in overhauling the security and resolute nature of physiological-based individual affirmation
systems like EMG.

Here is an illustration of the CNN architecture that (Naseer et al., 2022) used [12].

Fig. 2. Convolutional Neural Network (CNN) Architecture

2) RQ2: What amount do significant realizing strategies show suitability in the area of
finger impression affirmation near ordinary techniques, as shown by the
composition?
Page 9 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264
 Page 10 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264

Table 5. Fingerprint Recognition System Performance

Fingerprint AUC Standardized SEM CI
Recognition System AUC
Based on CNN 86.3 39.1574 0.00928 0.86276–0.86392
Based on NN 86.6 41.7123 0.00879 0.86590–0.86700
Classifier
Based on FL 83 28.7231 0.01148 0.82895–0.83039
Classifier
Based on LDA 83.2 13.4341 0.01470 0.69654–0.69838
Classifier
Considering the article (El-Rahman et al., "Enhanced multimodal biometric recognition
systems based on deep learning and traditional methods in smart environments" 2024) [13].
As shown in the table 5 the assessment comprehensively dove into the practicality of
significant learning strategies, particularly Convolutional Neural Networks (CNN), in the
space of finger impression affirmation versus standard techniques. Fundamental revelations
ascended out of this assessment: Above all else, to the extent that execution estimations,
CNN-based finger impression affirmation systems showed essential AUC (Area Under the
Curve) values going from 83% to 86.6%, beating various classifiers like Neural Networks
(NN), Fuzzy Logic (FL), and Linear Discriminant Analysis (LDA). In addition, in the space
of multimodal biometric structures joining finger impression and Electrocardiogram (ECG)
modalities, CNN-based blend methods displayed transcendent execution, with AUC (Area
Under the Curve) values outperforming 98.1%. These results feature the huge advantages of
CNN in working on the precision and capability of remarkable finger impression affirmation
structures, proposing promising opportunities for supporting data approval and access control
frameworks across various endeavors and applications. Additionally, the appraisal includes
the sensibility of critical learning perspectives, especially CNN, in both unimodal and
multimodal biometric structures. Unmistakably, CNN-based mix strategies in multimodal
frameworks beat standard philosophies, showing higher accuracy and ability. The
relationship with standard ways of thinking revealed the inescapability of CNN-based
structures, as avowed by dominating execution assessments like AUC (Area Under the
Curve), attitude, responsiveness, and capacity. All around, these revelations supplement the
capacity of significant learning systems, especially CNN, in driving the field of exceptional
finger impression affirmation and reinforcing data security in various valuable
circumstances, thus supporting their wide gathering in future biometric approval structures.

3) RQ3: How do variations in dataset characteristics impact the performance of deep

learning models for fingerprint recognition?
Considering the article (Althnian et al., "Impact of dataset size on classification
performance: An empirical evaluation in the medical domain" 2021) [14]. Table 6 illustrates
the intricate relationship between dataset variations and the performance of deep learning
models in fingerprint recognition tasks.

Table 6. Table Data Synthetic

Dataset Variation Impact on Deep Learning Model Performance
Characteristic

Size of Datasets Small (Consist of 1,000 samples) Low precision, overfitting

Size of Datasets Medium (Consist of 10,000 samples) limited generalizability and moderate accuracy

Page 10 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264

 Page 11 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264

Size of Datasets Large (Consist of 100,000 samples) High precision, great generalizability

Dataset Low (samples taken from one Poor generalizability and population bias of the
Diversity population only) model

Dataset Medium (samples taken from a few Less one-sided model, moderate generalizability
Diversity populaces)

Dataset High (samples taken from worldwide Model with no bias and high generalizability
Diversity populace)

Dataset Quality Low (a lot of noise and distortion) Low exactness, shaky model

Dataset Quality Medium (some clamor and antiquities) Moderate exactness, semi-stable model

The size, diversity, and quality of a dataset all significantly impact the performance of
deep learning models for fingerprint recognition. Larger datasets typically lead to higher
accuracy as the model has more information to learn from. However, this can come at the
cost of generalizability if the data lacks diversity. Diverse datasets, encompassing a wider
range of fingerprints, improve the model's ability to perform well on unseen data. On the
other hand, high levels of noise and artifacts in the dataset can hinder both accuracy and
generalizability. To address these quality issues, data preprocessing techniques like noise
removal and standardization can be applied before training the model.

4. CONCLUSION
This study provides a comprehensive overview of significant learning in unique
fingerprint authentication by focusing on articles published between 2019 and 2024. Through
careful chase, assessment, and combination, key models and disclosures are highlighted,
highlighting the sufficiency of Convolutional Neural Networks (CNNs), LSTM-RNN, and
Support Vector Machines (SVMs) in watching out for affirmation challenges. With well-
known papers demonstrating clear goals, robust evaluation strategies, and comprehensive
discoveries, quality evaluation confirms adherence to key norms. Certainly, CNN-based
frameworks perform exceptionally well, with AUC values ranging from 83% to 86.6%,
particularly when dealing with complex examples such as EMG signals. Additionally, our
investigation demonstrates how the model's performance is affected by the characteristics of
the dataset. This proposes that bigger and more differed datasets improve generalizability
and precision. For instance, datasets with 100,000 samples exhibit excellent generalizability
and precision. However, the accuracy and soundness of datasets with noise and mutilation
decrease. These discoveries underline the meaning of integrating broad learning strategies
into impending biometric affirmation systems to improve accuracy, proficiency, and security
across different applications and settings.

REFERENCES
[1] J. Galbally, F. Alonso-Fernandez, J. Fierrez and J. Ortega-Garcia, “A High-Performance
Fingerprint Liveness Detection Method Based on Quality-Related Features,” Future
Generation Computer Systems, vol. 28, no. 1, pp. 311-321, 2012.

[2] M. A. M. Abdullah, R. R. Al-Nima, S. S. Dlay, W. L. Woo and J. A. Chambers, “Cross-

spectral Iris Matching for Surveillance Applications,” Springer, Surveillance in Action
Technologies for Civilian, Military and Cyber Surveillance, Chapter 5, 2017.
Page 11 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264
 Page 12 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264

[3] M. R. Khalil, M. S. Majeed, and R. R. Omar, “Personal identification with iris patterns,”
AL-Rafidain Journal of Computer Sciences and Mathematics, vol. 6, no. 1, pp.13-26, 2009.

[4] S. A. M. Al-Sumaidaee, M. A. M. Abdullah, R. R. O. Al-Nima, S. S. Dlay and J. A.

Chambers, “Multi-gradient Features and Elongated Quinary Pattern Encoding for Image-
based Facial Expression Recognition,” Elsevier, Pattern Recognition, vol. 71, pp. 249 - 263,
2017.

[5] R. R. Al-Nima, F. S. Mustafa, “Face Recognition Using Invariant Moments Features,”

The Scientific Journal, College of Science, Tikrit University, vol. 14, no. 2, pp. 253-259,
2009.

[6] R. Alimuin, E. Dadios, J. Dayao, and S. Arenas, “Deep hypersphere embedding for real-
time face recognition,” TELKOMNIKA Telecommunication Computing Electronics and
Control, vol. 18, no. 3, pp. 1671-1677, 2020.

[7] R. R. O. Al-Nima, M. Y. Al-Ridha, F. H. Abdulraheem, “Regenerating face images from

multi-spectral palm images using multiple fusion methods,” TELKOMNIKA
Telecommunication Computing Electronics and Control, vol. 17,no. 6, pp. 3110-3119, 2019.

[8] A. Ramadhan, “Data Capital: A Systematic Literature Review,” DESIDOC Journal of

Library and Information Technology, vol. 42, no. 2. 2022. doi: 10.14429/djlit.42.2.17503

[9] R. Ryan, “Cochrane Consumers and Communication Review Group: data synthesis and
analysis,” Cochrane Consumers and Communication Review Group, vol. 2013, no. June
2013.

[10] V. Kamble, M. Dale, and V. Bairagi, “A Hybrid Model by Combining Discrete Cosine
Transform and Deep Learning for Children Fingerprint Identification,” International Journal
of Advanced Computer Science and Applications, vol. 14, no. 1, 2023, doi:
10.14569/IJACSA.2023.0140186.

[11] B. BİLGİN, M. İ. GÜRSOY, and A. ALKAN, “Biometric Personal Classification with

Deep Learning Using EMG Signals,” Bilge International Journal of Science and Technology
Research, vol. 7, no. 2, 2023, doi: 10.30516/bilgesci.1344337.

[12] I. Naseer, S. Akram, T. Masood, A. Jaffar, M. A. Khan, and A. Mosavi, “Performance

Analysis of State-of-the-Art CNN Architectures for LUNA16,” Sensors, vol. 22, no. 12,
2022, doi: 10.3390/s22124426.

[13] S. A. El-Rahman and A. S. Alluhaidan, “Enhanced multimodal biometric recognition

systems based on deep learning and traditional methods in smart environments,” PLoS One,
vol. 19, no. 2 February, 2024, doi: 10.1371/journal.pone.0291084.

[14] A. Althnian et al., “Impact of dataset size on classification performance: An empirical

evaluation in the medical domain,” Applied Sciences (Switzerland), vol. 11, no. 2, 2021, doi:
10.3390/app11020796.

Page 12 of 12 - AI Writing Submission Submission ID trn:oid:::1:2903219264