Journal of Network Security Computer Networks

Brain Signal Security and Ethical Considerations in ML/DL-based BCIs

R. Naveenkumar — 2025-11-06

Machine Learning (ML) and Deep Learning (DL)-based Brain–Computer Interfaces (BCIs) have dramatically transformed human-technology interaction, enabling applications that range from assistive technologies to neurorehabilitation. The application of ML/DL-based BCIs brings serious concerns related to brain signal security and ethics. These are brain signals; hence, by nature, electroencephalography (EEG) and functional Magnetic Resonance Imaging (fMRI) can reveal personal information regarding an individual’s mental state, intentions, and health. Therefore, unauthorized access, manipulation, or misuse of such data poses serious risks, such as privacy breaches, identity theft, and malicious control over neuroprosthetics. This paper investigates a set of critical security vulnerabilities in ML/DL-based BCIs, adversarial attacks, data poisoning, and model inversion, which may simultaneously violate the integrity and confidentiality of the brain data. If this is so, then ethical considerations include informed consent, ownership of data, and fair access to BCI technologies. The paper argues for incorporating robust encryption methods, secure data transmission protocols, and adversarial training of BCIs to protect against emerging threats. It also outlines an ethical compliance framework that focuses on transparency, accountability, and adherence to human rights principles. In this context, the challenges presented above must be addressed as BCI technology is translated from the research environment into widespread application to ensure safe, fair, and trustworthy integration into society. This work seeks to stimulate debate among researchers, practitioners, and policymakers in pushing for interdisciplinary approaches to balance innovation with the demands of ethics and security in this transformative technology.

Using Deep Learning for Behavioral Pattern Recognition for Avoiding Cyberbullying: A Comprehensive Review

Manas Kumar Yogi — 2025-10-16

Cyberbullying poses severe psychological and emotional risks worldwide. Traditional monitoring and rule-based detection methods struggle with scalability and contextual adaptability. This review critically examines deep learning-based behavioral pattern recognition models as a scalable and adaptive solution for cyberbullying detection. Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM), and Transformer models enable fine-grained feature extraction from multimodal social data, effectively capturing linguistic, temporal, and visual cues of online aggression. The term “psychometrically validated” is replaced with “empirically validated through benchmarked datasets,” ensuring methodological transparency. Comparative analyses of recent deep learning studies demonstrate accuracy improvements of 5–12% over traditional models when applied to benchmark datasets such as Kaggle’s “Cyberbullying Detection” and Twitter Hate Speech datasets. Key challenges, including data imbalance, interpretability, and privacy, are linked to emerging research directions involving fairness-aware, explainable, and privacy-preserving deep architectures.

Design and Evaluation of CryptoKen: A Tokenomics-driven Ethereum Cryptocurrency with Governance and Interoperability Features

Serena Leothes Gomez — 2025-11-10

CryptoKen is presented as an ERC-20 governance token that utilizes quadratic voting to increase engagement in decentralized applications (dApps) and overcome the governance challenges, including voter disengagement, of most token-based systems. The token was developed in Solidity ^0.8.20 and evaluated through unit and integration testing with Foundry, static analysis using Slither v0.10.0, fuzz testing with Echidna v2.1.0, user acceptance testing (UAT) involving 22 participants, and gas consumption benchmarking on a forked Sepolia testnet. The evaluation achieved 92% statement coverage (solidity-coverage v0.2.5; GitHub Actions run #123), successful execution of all 35 integration test scenarios, and no critical vulnerabilities reported in static analysis. UAT results indicated high usability, with a median System Usability Scale (SUS) score of 82.5 (IQR: 75–87.5). The outcomes highlight a reproducible methodology for assessing governance tokens, demonstrating that CryptoKen’s quadratic voting mechanism and simplified interface effectively address usability and technical challenges while providing a scalable foundation for advancing Decentralized Autonomous Organization (DAO) systems.

Network Traffic Intrusion Detection Applications: Key Parameters and Techniques

Twinkal Yadav — 2025-09-17

Intrusion Detection Systems (IDS) play a critical role in safeguarding digital infrastructures by identifying and mitigating unauthorized access and malicious activities. This paper presents a comprehensive overview of the various IDS techniques, including signature-based, anomaly-Intrusion Detection Systems (IDS) play a vital role in protecting digital infrastructures by detecting and preventing unauthorized access and malicious activities. This paper provides a comprehensive overview of various IDS techniques, including signature-based, anomaly-based, specification-based, and modern machine learning and deep learning methods. Each approach is assessed based on its decision-making processes, input needs, detection abilities, and performance metrics. The study also examines hybrid and heuristic models that aim to improve detection accuracy by combining the strengths of multiple techniques. A detailed comparison is included in tabular form, outlining the advantages, disadvantages, and limitations of each method. While traditional techniques offer high accuracy for known threats, they often fail to detect new and emerging ones. Conversely, learning-based models are effective at identifying unknown intrusions but face challenges like high false positive rates and increased computational requirements. This analysis serves as a valuable reference for researchers and practitioners to choose or develop suitable IDS strategies based on system needs, data features, and the evolving threat landscape. Based on specification and modern machine learning and deep learning approaches, each method is evaluated in terms of its decision-making patterns, input requirements, detection capabilities, and performance metrics. The study further explores hybrid and heuristic-based models that aim to enhance detection accuracy by leveraging the strengths of multiple techniques. A detailed comparative analysis is provided in tabular form, highlighting the advantages, disadvantages, and limitations of each approach. While traditional methods offer high precision for known attacks, they struggle with detecting novel threats. Conversely, learning-based models excel at identifying unknown intrusions but face challenges such as high false positives and computational overhead. This analysis offers a foundational reference for researchers and practitioners to select or design appropriate IDS strategies based on system requirements, data characteristics, and threat landscapes.