IET Information Security最新文献

The rapid proliferation of Internet of Things (IoT) devices has revolutionized various industries by enabling smart grids, smart cities, and other applications that rely on seamless connectivity and real-time data processing. However, this growth has also introduced significant security challenges due to the scale, heterogeneity, and resource constraints of IoT systems. Traditional intrusion detection systems (IDS) often struggle to address these challenges effectively, as they require centralized data collection and processing, which raises concerns about data privacy, communication overhead, and scalability. To address these issues, this paper investigates the application of federated learning for network intrusion detection in IoT environments. We first evaluate a range of machine learning (ML) and deep learning (DL) models, finding that the random forest model achieves the highest classification accuracy. We then propose a federated learning approach that allows distributed IoT devices to collaboratively train ML models without sharing raw data, thereby preserving privacy and reducing communication costs. Experimental results using the UNSW-NB15 dataset demonstrate that this approach achieves promising outcomes in the IoT context, with minimal performance degradation compared to centralized learning. Our findings highlight the potential of federated learning as an effective, decentralized solution for network intrusion detection in IoT environments, addressing critical challenges, such as data privacy, heterogeneity, and scalability.

Blockchain technology has reshaped numerous industries by providing secure and transparent transactional platforms. This paper delves into the intersection of blockchain analytics and artificial intelligence (AI) to advance transaction analysis. The primary aim is to bolster fraud detection and enhance transaction efficiency. Through a comprehensive literature review, we identify gaps in existing knowledge and lay the groundwork for our research. We introduce a novel transaction-hybrid model developed using machine learning (ML) algorithms, including support vector machines (SVMs), K-nearest neighbors (KNNs), and random forest (RF). This transact-hybrid model aims to fortify fraud detection capabilities by harnessing the strengths of each algorithm. We curate a unique dataset comprising 1000 instances, incorporating critical transaction features such as transaction hash, block number, transaction fee and gas limit, with binary classification indicating fraudulent transactions. Meticulous preprocessing, including feature engineering and data splitting for training and testing, is conducted. Visualization techniques, including seaborn-based graphs, correlation plots and violin plots, elucidate the dataset’s characteristics. Additionally, a spring colormap correlation map enhances the understanding of feature relationships. Transaction fee distributions before and after preprocessing are visually presented, highlighting the impact of data preparation. We introduce the novel transact-hybrid classifier (THC) with detailed mathematical equations, emphasising its contribution to transactional fraud detection. The classifier integrates SVM, KNN and RF outputs using an exclusive OR operation, showcasing innovation in model development. To evaluate model performance, we conduct a comparative analysis, incorporating SVM, KNN, RF and a voting classifier. Bar plots for accuracy, precision, recall and F1 score, with a custom plasma colormap, offer a visual summary of each model’s metrics. Furthermore, a receiver operating characteristics (ROC) curve analysis is presented, highlighting the area under the curve (AUC) for SVM, KNN, RF and voting models, providing a comprehensive view of their performance in distinguishing between true positive and false positive rates. Our proposed method demonstrates over 99% efficacy in fraud detection, underscoring its potential impact in transaction analysis.

Pseudorandom functions (PRFs) are a very important tool in cryptography, and the learning with rounding (LWR) problem is one of the main issues in their construction. LWR problem, is to find $$ from ⌊As⌋_p, where $$ and $$ is the rounding function. The LWR problem is considered a variant of the learning with error (LWE) problem, that is, to find s from b = As + e, where $$, and LWE has been reduced to GapSVP and SIVP. The hardness of the lattice problems is the security foundation of the issued schemes. The best-known reduction for LWR was completed using information-theoretic entropy arguments, and the reduction requires q ≥ 2nmp. It does not directly reduce to the closest vector problem (CVP) problem, but rather to the LWE problem. However, the reduction in the aforementioned work significantly reduces the difficulty of LWR. To more accurately characterize the hardness of LWR, this paper uses statistical approximation and a Quantum Fourier Transform to reduce LWR to the CVP, thereby ensuring the hardness of LWR. Furthermore, unlike the previous conclusions, our reduction involves minimal loss and has broad security conditions, requiring only that $$, where q and p are prime numbers and 0 < α < 1.

Background: The Internet of Things (IoT) represents one of the fastest-expanding developments in the computer industry. However, the inherently hostile environment of the internet makes IoT systems vulnerable. A popular and promising method for detecting cyberattacks is machine learning (ML), which produces excellent outcomes for identified attacks. However, their ability to identify unidentified malicious traffic is nearly nonexistent.

Need for the Study: The need for study arises from the advanced security solutions of IoT, which are vulnerable to various known and unknown cyberattacks. Traditional ML methods are used to effectively detect new threats. It is followed by a hybrid methodological framework to combine supervised and semisupervised learning. It is an advanced approach to enhance detection accuracy and adaptability in dynamic IoT environments.

Methods: The study suggests an innovative strategy that combines supervised and unsupervised techniques. Initially employing several flow-based parameters, the improved density-based spatial clustering of applications with noise (IDBSCAN) clustering technique distinguishes between anomalous and regular traffic. Next, utilizing specific statistical metrics, a hybrid multiple kernel extreme learning machine with modified teaching–learning-based optimization (HMKELM-MTLBO) classification process is applied to label the clusters.

Findings of the Study: The findings of accuracy result as 98.95%, precision as 97.65%, recall as 98.56%, and F1 score value as 98.23%.

Results: The approach’s effectiveness was evaluated using the ToN_IoT dataset, and a 99%+ accuracy rate was attained in identifying cyberattacks across IoT technology.

Conclusion: The study validates the suggested strategy by testing a distinct set of attacks and training on the ToN_IoT dataset utilizing an extensive data processing system.

There are user privacy risks in cloud-based vehicle dispatch platforms due to the unauthorized collection, use, and dissemination of data. However, existing data protection methods cannot balance privacy, usability, and efficiency well. To address this, we propose a local privacy-preserving vehicle assignment strategy via spatial–temporal fusion (STF-LPPVA). Specifically, the strategy allows the cloud platform to train and distribute a spatial–temporal representation model to the user side. Encoded by this model, drivers and passengers can privately fuze the spatial–temporal information of their trips and then transmit these fuzed vectors to the cloud platform. Based on the similarity of the vectors, the cloud platform can allocate vehicles using the Kuhn–Monkreth (KM) algorithm. In addition, we analyze the theoretical feasibility of the STF-LPPVA strategy using entropy change and get good performance with a dataset from DiDi in Chengdu, China. The results show that the successful matching rate of the STF-LPPVA strategy is very close to the original data matching with lower time overhead. Our approach can reduce the traveling distance by 66.5% and improve the matching success rate by 36.2% on average.

Recent advancements in unmanned aerial vehicle (UAV) technology have facilitated its widespread adoption across a spectrum of sectors, such as commercial logistics, agricultural surveillance, industrial diagnostics, and military maneuvers. However, the widespread adoption has also engendered a burgeoning array of security concerns. Unmanned aerial systems (UAS) networks are characterized by high node mobility, unstable links, open communication environments, and limited platform resources, which in turn exhibit typical vulnerabilities in terms of cybersecurity. Most current studies on UAV cybersecurity issues tend to focus on individual UAVs, often neglecting the holistic cybersecurity of UAS. This paper outlines the composition of UAS network architecture. It summarizes the main cybersecurity challenges UAS faces within six categories—spoofing, tampering, information disclosure, denial of service (DoS), service refusal, and privilege escalation—based on the STRIDE threat model. Corresponding methods for risk mitigation and security protection strategies are proposed. Ultimately, the paper provides a perspective on the future development directions of UAS cybersecurity, aiming to offer a reference for addressing related issues in subsequent research and practice.

The COVID-19 pandemic has impacted the world, prompting a shift toward remote work and stay-at-home economies, altering routines for individuals and businesses. Organizations have had to swiftly implement digital solutions to enable productive and efficient remote work, a trend that is becoming increasingly common. In this context, enterprise programmers often rely on open-source software from social platforms to accelerate application development. However, the source code on these platforms may not always be regularly updated or well-maintained, posing security risks. These risks are exacerbated when programmers need more security software-focused development practices, testing for vulnerabilities, or applying necessary patches regularly. This study introduces two secure software development (SSD) performance baselines based on international standards and utilizing statistical process control (SPC): proactive information security awareness and reactive risk management. These baselines enable enterprise IT departments to monitor security awareness and improve the secure development capabilities of programmers and R&D teams, thereby mitigating the security risks of released software. A practical case study is presented to demonstrate the effectiveness of this approach.

Blockchain technology has become a popular choice for electronic voting systems due to its transparency, security, and decentralization. However, it is not a perfect solution, as its inherent immutability poses challenges in blockchain-based e-voting systems. Specifically, without the physical security provided by traditional polling stations, preventing bribery and coercion becomes more difficult. Additionally, because of blockchain’s immutability, voters who are coerced or mistakenly vote cannot correct their choice. To address these issues, this paper proposes a secure blockchain-based voting system with editable ballots. The system uses chameleon hashes with ephemeral trapdoors and a timestamp mechanism, allowing voters to modify their ballots within a legitimate timeframe. Additionally, a modified Paillier cryptosystem and blind signature technology are used to ensure that any modifications leave no trace. We simulate and evaluate the system using Fabric 2.2, focusing on computational complexity and system stability. Analysis of experimental results shows that the blockchain-based voting system with an editable ballot mechanism proposed in this article has good computational cost and stability performance under normal use pressure.

Reversible data hiding in encrypted images (RDHEI) is a technique that not only allows the cover images can be fully restored without any loss of information after the embedded data has been extracted but also ensures the confidentiality within the cover images. This article proposes an RDHEI scheme combining adaptive (n, n) secret image sharing (SIS) manner. The content owner reserves part of the least significant bit plane (LSBP) in cover images by two most significant bit planes (MSBPs) compression using the median edge detector (MED) prediction method. To level up the privacy protection of n cover images, a two-layer encryption method is utilized to generate n shares, that is, the self-encryption and cross-encryption. Moreover, our method can be applied on no matter how many of cover images. The secret data with identification can be concealed by the data hiders into the vacated LSB of their own shares. Through the cooperation of the overall shares, the receiver can retrieve the embedded secret data and recover the cover images. Experiment results reveal the security reliability of our approach and the outstanding performance when compared to some related methods. Also, the approach can be employed in color image domain.

The current multikey fully homomorphic encryption (MKFHE) needs to add exponential noise in the distributed decryption phase to ensure the simulatability of partial decryption. Such a large noise causes the ciphertext modulus of the scheme to increase exponentially compared to the single-key fully homomorphic encryption (FHE), further reducing the efficiency of the scheme and making the hardness problem on the lattice on which the scheme relies have a subexponential approximation factor $$ (which means that the security of the scheme is reduced). To address this problem, this paper analyzes in detail the noise in partial decryption of the MKFHE based on the learning with error (LWE) problem. It points out that this part of the noise is composed of private key and the noise in initial ciphertext. Therefore, as long as the encryption scheme is leak-resistant and the noise in partial decryption is independent of the noise in the initial ciphertext, the semantic security of the ciphertext can be guaranteed. In order to make the noise in the initial ciphertext independent of the noise in the partial decryption, this paper proves the smudging lemma on discrete Gaussian distribution and achieves this goal by multiplying the initial ciphertext by a “dummy” ciphertext with a plaintext of 1. Based on the above method, this paper removes the exponential noise in the distributed decryption phase for the first time and reduces the ciphertext modulus of MKFHE from 2^{ω(λL logλ)} to 2^{O(λ + L)} as the same level as the FHE.