Network and cybersecurity applications of defense in adversarial attacks: A state-of-the-art using machine learning and deep learning methods

Journal of Intelligent Systems, Journal Year: 2024, Volume and Issue: 33(1)

Published: Jan. 1, 2024

Abstract This study aims to perform a thorough systematic review investigating and synthesizing existing research on defense strategies methodologies in adversarial attacks using machine learning (ML) deep methods. A methodology was conducted guarantee literature analysis of the studies sources such as ScienceDirect, Scopus, IEEE Xplore, Web Science. question shaped retrieve articles published from 2019 April 2024, which ultimately produced total 704 papers. rigorous screening, deduplication, matching inclusion exclusion criteria were followed, hence 42 included quantitative synthesis. The considered papers categorized into coherent classification including three categories: security enhancement techniques, attack mechanisms, innovative mechanisms solutions. In this article, we have presented comprehensive earlier opened door potential future by discussing depth four challenges motivations attacks, while recommendations been discussed. science mapping also performed reorganize summarize results address issues trustworthiness. Moreover, covers large variety network cybersecurity applications subjects, intrusion detection systems, anomaly detection, ML-based defenses, cryptographic techniques. relevant conclusions well demonstrate what achieved against attacks. addition, revealed few emerging tendencies deficiencies area be remedied through better more dependable mitigation methods advanced persistent threats. findings crucial implications for community researchers, practitioners, policy makers artificial intelligence applications.

Language: Английский

---

Adversarial Attacks in Machine Learning: Key Insights and Defense Approaches

Applied Data Science and Analysis, Journal Year: 2024, Volume and Issue: 2024, P. 121 - 147

Published: Aug. 7, 2024

There is a considerable threat present in genres such as machine learning due to adversarial attacks which include purposely feeding the system with data that will alter decision region. These are committed presenting different models way model would be wrong its classification or prediction. The field of study still relatively young and has develop strong bodies scientific research eliminate gaps current knowledge. This paper provides literature review defenses based on highly cited articles conference published Scopus database. Through assessment 128 systematic articles: 80 original papers 48 till May 15, 2024, this categorizes reviews from domains, Graph Neural Networks, Deep Learning Models for IoT Systems, others. posits findings identified metrics, citation analysis, contributions these studies while suggesting area’s further development robustness’ protection mechanisms. objective work basic background defenses, need maintaining adaptability platforms. In context, contribute building efficient sustainable mechanisms AI applications various industries

Language: Английский

---

Fuzzy Evaluation and Benchmarking Framework for Robust Machine Learning Model in Real-Time Autism Triage Applications

International Journal of Computational Intelligence Systems, Journal Year: 2024, Volume and Issue: 17(1)

Published: June 17, 2024

Abstract In the context of autism spectrum disorder (ASD) triage, robustness machine learning (ML) models is a paramount concern. Ensuring ML faces issues such as model selection, criterion importance, trade-offs, and conflicts in evaluation benchmarking models. Furthermore, development must contend with two real-time scenarios: normal tests adversarial attack cases. This study addresses this challenge by integrating three key phases that bridge domains fuzzy multicriteria decision-making (MCDM). First, utilized dataset comprises authentic information, encompassing 19 medical sociodemographic features from 1296 autistic patients who received diagnoses via intelligent triage method. These were categorized into one labels: urgent, moderate, or minor. We employ principal component analysis (PCA) algorithms to fuse large number features. Second, fused forms basis for rigorously testing eight models, considering scenarios, evaluating classifier performance using nine metrics. The third phase developed robust framework encompasses creation decision matrix (DM) 2-tuple linguistic Fermatean opinion score method (2TLFFDOSM) multiple-ML perspectives, accomplished through individual external group aggregation ranks. Our findings highlight effectiveness PCA algorithms, yielding 12 components acceptable variance. ranking, logistic regression (LR) emerged top-performing terms 2TLFFDOSM (1.3370). A comparative five benchmark studies demonstrated superior our across all six checklist comparison points.

Language: Английский

---

Optimal Time Window Selection in the Wavelet Signal Domain for Brain–Computer Interfaces in Wheelchair Steering Control

Applied Data Science and Analysis, Journal Year: 2024, Volume and Issue: 2024, P. 69 - 81

Published: June 15, 2024

Background and objective: Principally, the procedure of pattern recognition in terms segmentation plays a significant role BCI-based wheelchair control system for avoiding errors, which can lead to initiation wrong command that will put user unsafe situations. Arguably, each subject might have different motor-imagery signal powers at times trial because he or she could start (or end) performing task slightly time intervals due differences complexities his her brain. Therefore, primary goal this research is develop generic model (GPRM)-based EEG-MI brain-computer interface steering control. Additionally, having simplified well generalized essential based BCI applications. Methods: Initially, bandpass filtering using multiple windows were used denoising finding best duration contains MI feature components. Then, extraction was performed five statistical features, namely minimum, maximum, mean, median, standard deviation, extracting components from wavelet coefficient. seven machine learning methods adopted evaluated find classifiers. Results: The results study showed that, durations time-frequency domain range (4-7 s). Interestingly, GPRM on LR classifier highly accurate, achieved an impressive classification accuracy 85.7%.

Language: Английский

---

A Frequency-Domain Pattern Recognition Model for Motor Imagery-Based Brain-Computer Interface

Applied Data Science and Analysis, Journal Year: 2024, Volume and Issue: 2024, P. 82 - 100

Published: June 20, 2024

Brain-computer interface (BCI) is an appropriate technique for totally paralyzed people with a healthy brain. BCI based motor imagery (MI) common approach and widely used in neuroscience, rehabilitation engineering, as well wheelchair control. In control system the procedure of pattern recognition term preprocessing, feature extraction, classification plays significant role performance. Otherwise, errors can lead to wrong command that will put user unsafe conditions. The main objectives this study are develop generic model-based EEG –MI interfaces steering signal filtering, segmentation, multiple time window was de-noising finding MI feedback. five statistical features namely (mean, median, min, max, standard deviation) were extracting frequency domain. classification, seven machine learning towards single hybrid classifier model. For validation, data from Competition dataset (Graz University) validate developed obtained result following: (1) preprocessing perspective it seen two-second optimal (2) have good efficiency EEG-MI (3) Classification using (MLP-LR) perfect domain Finally, be concluded efficient deployed real-time system.

Language: Английский

---

Prioritizing disability support systems by using Tamir’s complex fuzzy Dombi aggregation operators

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: May 20, 2025

Language: Английский

---

Semantic Image Retrieval Analysis Based on Deep Learning and Singular Value Decomposition

Applied Data Science and Analysis, Journal Year: 2024, Volume and Issue: 2024, P. 17 - 31

Published: March 25, 2024

The exponential growth in the total quantity of digital images has necessitated development systems that are capable retrieving these images. Content-based image retrieval is a technique used to get from database. user provides query image, and system retrieves those photos database most similar image. problem pertains task locating photographs inside extensive datasets. Image researchers transitioning use keywords utilization low-level characteristics semantic features. push for features arises issue subjective time-consuming keywords, as well limitation capturing high-level concepts users have mind. main goal this study examine how convolutional neural networks can be acquire advanced visual These feature descriptors potential effective compared handcrafted terms representation, which would result improved performance. (CBIR-VGGSVD) model an ideal solution content-based based on VGG-16 algorithm uses Singular Value Decomposition (SVD) technique. suggested incorporates purpose extracting both kept Afterwards, dimensionality retrieved reduced using SVD. Then, we compare dataset cosine metric see they are. When all said done, share high degree similarity will successfully extracted dataset. A validation performance CBIR-VGGSVD performed Corel-1K standard sole one used, implementation produce average precision 0.864. On other hand, when utilized, revealed (0.948). findings ensured provided improvement test pictures were surpassing recent approaches.

Language: Английский

---

Generalized Time Domain Prediction Model for Motor Imagery-based Wheelchair Movement Control

Mesopotamian Journal of Big Data, Journal Year: 2024, Volume and Issue: 2024, P. 68 - 81

Published: June 15, 2024

Brain-computer interface (BCI-MI)-based wheelchair control is, in principle, an appropriate method for completely paralyzed people with a healthy brain. In BCI-based system, pattern recognition terms of preprocessing, feature extraction, and classification plays significant role avoiding errors, which can lead to the initiation wrong command that will put user unsafe condition. Therefore, this research's goal is create time-domain generic model (GPRM) two-class EEG-MI signals use system.This GPRM has advantage having applicable unknown subjects, not just one. This been developed, evaluated, validated by utilizing two datasets, namely, BCI Competition IV Emotive EPOC datasets. Initially, fifteen-time windows were investigated seven machine learning methods determine optimal time window as well best strong generalizability. Evidently, experimental results study revealed duration signal range 4–6 seconds (4–6 s) high impact on accuracy while extracting features using five statistical methods. Additionally, demonstrate one-second latency after each cue when eight-second Graz protocol used study. inevitable because it practically impossible subjects imagine their MI hand movement instantly. at least one second required prepare initiate motor imagery movement. Practically, are efficient viable decoding domain. based LR classifier showed its ability achieve impressive 90%, was dataset. The developed highly adaptable recommended deployment real-time EEG-MI-based systems.

Language: Английский

---

Optimizing pool mining performance: A VIKOR‐based model for identifying reputed miners in blockchain networks

Concurrency and Computation Practice and Experience, Journal Year: 2024, Volume and Issue: 36(21)

Published: June 23, 2024

Abstract Blockchain networks continue to gain attraction in cutting‐edge applications and mining within these has become increasingly popular. To get rewards, miners solve cryptographic puzzles add new blocks blockchain using the proof‐of‐work (PoW) consensus mechanism. Numerous opt participate pools due challenges of solo mining. However, selecting reputed for pool poses a significant challenge, given decentralized nature system. This paper addresses this challenge by introducing ranking model that evaluates miners' performance reputation through trust scores. It provides method optimizing identifying highly pools, enhancing overall profitability. endeavor necessitates development algorithms tailored unique dynamics pools. The research offers meticulously designed identifies miners. We extensively evaluate proposed hyperledger framework, guaranteeing strong across vital metrics like block authorization time, Processing creation validation confirmation time.

Language: Английский

---

Deep Transfer Learning Model for EEG Biometric Decoding

Applied Data Science and Analysis, Journal Year: 2024, Volume and Issue: 2024, P. 4 - 16

Published: Feb. 28, 2024

In automated systems, biometric systems can be used for efficient and unique identification authentication of individuals without requiring users to carry or remember any physical tokens passwords. Biometric are a rapidly developing promising technology domain. in contrasting with conventional methods like password IDs. Biometrics refer biological measures traits that employed identify authenticate individuals. The motivation employ brain activity as identifier automatic has increased substantially recent years. specific focus on data obtained through electroencephalography (EEG). Numerous investigations have revealed the existence discriminative characteristics signals captured during different types cognitive tasks. However, because their high dimensional nonstationary properties, EEG inherently complex, which means both feature extraction classification must take this into consideration. study, hybridization method combined classical classifier pre-trained convolutional neural network (CNN) short-time Fourier transform (STFT) spectrum was employed. For tasks such subject lock unlock classification, we hybrid model mobile decode two-class motor imagery (MI) signals. This accomplished by building nine distinct models using potential classifiers, primarily algorithms, from best one finally selected. experimental portion study involved, practice, six experiments. tasks, first experiment tries create model. order accomplish this, were constructed largely methods. Comparing RF-VGG19 other models, it is evident former performed better. As result, chosen authentication. performance validated second experiment. third attempts verifying model's performance. fourth performs process an average accuracy 91.0% fifth verify effectiveness performing task. mean achieved 94.40%. Validating task dataset (unseen data) goal sixth experiment, 92.8%. indicates assesses left right hands' ability MI signal. Consequently, aid BCI-MI community simplifying implementation requirement, specifically classification.

Language: Английский

---