Blockchain technology has emerged as a transformative tool for secure data sharing across decentralised systems, particularly in finance, healthcare, and governance. However, despite its promise, the widespread adoption of blockchain platforms remains constrained by unresolved security threats and architecture-specific vulnerabilities. This paper presents a systematic literature review (SLR) that critically evaluates the security risks and countermeasures associated with blockchain-based data sharing models. The review focuses on three widely referenced platforms: Ethereum, Hyperledger Fabric, and MedRec, which are chosen for their relevance to public, permissioned, and healthcare-oriented blockchain deployments, respectively. The review analyzed 30 peer-reviewed publications from 2018 to 2025 sourced from IEEE Xplore, SpringerLink, ScienceDirect, and other digital libraries. Empirical insights from the reviewed literature indicate that Sybil attacks remain prevalent on public blockchains, although adaptive Proof-of-Stake protocols are reported to reduce their success rate considerably. Front-running and Miner Extractable Value–related behaviors are frequently observed in Ethereum-based decentralised finance ecosystems, often resulting in significant financial losses. Unauthorised access persists as a major concern, particularly for software wallets, which are commonly exposed to phishing and malware attacks. The findings underscore unique trade-offs across platforms: Ethereum supports transparency but is prone to transaction manipulation; Hyperledger ensures strong access control yet faces insider threat challenges; and MedRec enhances patient privacy but lacks robust mobile integration. This study provides a structured synthesis of existing threats, platform-level responses, and design trade-offs, offering guidance for stakeholders aiming to strengthen security in blockchain-based data-sharing infrastructures.

Large groups can make decisions via techniques like voting, referendums, and elections. During the realization and evaluation of these events, time efficiency, counting integrity, and voting reliability are crucial factors. Users can create their own polls, votes, and surveys using the interfaces created in this study. On these designed election processes, they can cast an electronic ballot. The CNN machine learning system evaluates the votes, or the counting process, with great accuracy, preventing manipulation and bias. Blockchain and zero-knowledge proof-based infrastructures support evaluation processes concurrently, enhancing voting privacy, data security, and transparency procedures. The results of the proposed CNN algorithm and the data privacy metrics demonstrate satisfactory performance.

Clustering in wireless sensor networks (WSNs) offers numerous desirable properties, including load balancing, energy conservation, and distributed key management. Secure Clustering requires it to detect compromised nodes and remove them from clusters during setup. If compromised nodes bypass the detection mechanism, they may disrupt the clustering process by altering cluster formations or initiating malicious clusters, thereby degrading overall network quality. To address these issues, a new method, Secretary Bird with Self-Organizing Maps (SBWSOM), has been designed to detect and eliminate malicious nodes while efficiently providing data. First, sensor nodes were deployed in a Python-based simulation environment. Second, malicious nodes were identified and eliminated, and the Cluster Head (CH) was selected based on parameters such as residual energy, distance to the base station (BS), and network topology. Furthermore, the data rates of the selected CHs were monitored, and data was transmitted to the sink node. Finally, performance metrics including latency, throughput, packet delivery ratio (PDR), energy consumption, and transmission loss were evaluated. The evaluation of this proposal demonstrated improved data transfer, with a throughput of 0.91, an energy consumption of 0.46 mJ, and a packet delivery ratio of 96.3%. Additionally, the transmission loss was 4.20%, and the latency was 6.04 ms. Overall, this method performed well, with significant improvement over previous models.

This paper presents a fully integrated, real-time assistive system that combines voice-based object recognition with a generative conversational interface, specifically designed to enhance elderly care via edge AI deployment. The proposed framework enables intuitive human–robot interaction in domestic environments by fusing natural language understanding, optimized visual detection, and on-device generative AI. Voice commands are processed via a speech-to-text pipeline using the Google Web Speech API. Keyword extraction triggers object detection using a quantized YOLOv8n model, accelerated by TensorRT with FP16 inference on an NVIDIA Jetson Nano. In parallel, a locally deployed generative AI assistant, executed entirely on-device, provides empathetic dialogue to support social engagement and emotional well-being. The proposed system adopts a hybrid edge architecture in which object detection, robot control, and LLM-based dialogue generation are executed on-device, while speech-to-text transcription relies on a cloud-based service. This generative interface is implemented as an LLM-based Emotional Assistant. The system achieves 13 FPS with an inference latency of 70 ms for object detection, 94.3% speech recognition accuracy, and an F1-score of 0.69 at a 0.5 confidence threshold. All AI components are executed on-board, preserving privacy for on-device processing while maintaining real-time responsiveness. Experimental validation confirms the effectiveness of deploying multimodal AI, including generative models, on resource-constrained hardware. This work lays the foundation for autonomous, voice-guided care robots that not only assist in locating objects but also engage users socially, promoting greater autonomy and quality of life for older adults.

The 5G-enabled Wireless Sensor Networks (WSNs) leverage the enhanced capabilities of 5G technology to evolve conventional WSN environments. WSN performance may be affected by interference from high-density 5G networks. The novel Hummingbird-based Graph Bernoulli Binomial Trust Management Network (HBbGBBTMN) proposed in this research aims to enhance smart, secure, and energy-efficient routing in 5G-enabled wireless networks. Python is initially used to simulate and model the network under consideration, accounting for fluctuating network conditions and dynamic node behaviors. An improved Hummingbird algorithm is used to detect and remove nodes with high energy consumption, thereby minimizing routing inefficiencies and avoiding suspicious behavior. To protect data integrity and prevent route disruption, malicious nodes are continuously detected and removed. The trust of the remaining nodes is calculated using the Bernoulli-Binomial distribution, which estimates each node's trust based on its previous packet-forwarding history. Such trust mechanisms are combined with node energy levels as well as network dynamics to form a fitness function that identifies optimal routing patterns. The proposed system is validated through extensive performance analysis, including measurements of packet delivery ratio, throughput, packet drop rate, delay, and malicious node prediction accuracy. The findings indicate that in decentralized wireless environments, HBbGBBTMN significantly enhances network efficiency, security, and dependability.

Depression is a serious psychiatric disorder that greatly impacts the quality of life and daily functioning of a person. Accurate diagnosis at the early stage is critical for success with intervention. Electroencephalography (EEG) offers a non-invasive technique to assess neurophysiological activity and is thus an important instrument for diagnosis of depression. Current EEG-based deep learning approaches are beset by high-dimensional data, poor feature selection, and poor classification performance owing to the nature of the EEG signal. To address these issues, we introduce EEGEffV2-SpikeNet, a new framework for depression detection that combines statistical feature extraction with deep feature extraction through a Graph Convolutional Network (GCN) approach. The proposed model incorporates a new fusion of statistical feature extraction and GCN-based deep feature learning for the extraction of both spatial and temporal EEG features. The extracted features are then optimized by the Modified Addax Optimization Algorithm (MAOA), a cutting-edge bio-inspired optimization algorithm that improves feature selection efficiency by discarding redundant information and enhancing classification accuracy. For depression classification, we utilize EfficientNetV2, Deep Belief Network (DBN), and Spiking Neural Network (SNN) to enhance feature representation and decision-making, leveraging the computational efficiency of EfficientNetV2 and the biologically plausible processing of SNN. Experimental results on two standard EEG datasets validate the better performance of the model, achieving 98.74% accuracy on Dataset 1 and 97.88% accuracy on Dataset 2, outperforming baseline models like DBN, EfficientNet, and SNN. The results prove the framework's promise as a dependable tool for objective and early depression diagnosis, with clinical application and mental health monitoring implications.

The growth of Internet of Things (IoT) networks has drastically expanded the attack surface, requiring intrusion detection systems (IDS) to ensure accuracy and privacy protection. To overcome these obstacles, we introduce a federated learning (FL)-based IDS that incorporates state-of-the-art preprocessing, smart feature optimization, and a new classification paradigm. During preprocessing, raw traffic data undergoes rigorous scrubbing, normalization via scaling, and label encoding to maintain consistency and reduce noise in heterogeneous local datasets. For feature selection, the Hybrid Emperor Penguin–Quokka Swarm Optimization (HEPQSO) approach is utilized which balances exploitation and exploration to find the most discriminative features while addressing the dimensionality problem. These features are then utilized by a deep hybrid classifier where the Spike Gated Linear Unit (SGLU) facilitates non-linear representation learning, and a Vision Transformer-Temporal Convolutional Network (ViT–TCN) hybrid discovers both global spatial relationships and local temporal dynamics of intrusion patterns. Experimental analyses performed using benchmark intrusion detection datasets show that the system consistently outperforms baseline models, with an accuracy of 97.88%, precision of 96.16%, recall of 97.54%, F1-score of 97.39%, specificity of 97.62%, and MCC of 97.04%, thus demonstrating its efficacy in secure IoT environments. This combination of state-of-the-art preprocessing, hybrid feature selection, and deep federated classification forms a robust IDS that can tackle the changing landscape of cyber intrusions.

In high-hazard workplaces like packaging facilities, effective fire safety is critical, but conventional practices often lack immersive realism and are costly to implement. This paper presents a hybrid reconstruction and artificial intelligence-driven framework that can potentially be applied to build interactive virtual reality environments. The objective of this study is to develop a scalable and cost-effective VR-based fire safety training system that balances realism and interactivity. To balance visual fidelity and interactivity, a hybrid reconstruction pipeline was developed. The complex background environment was reconstructed and rendered using 3D Gaussian Splatting, while photogrammetry was used to reconstruct key industrial objects as solid, interactive meshes. An AI-based system was employed for automatic object detection using You Only Look Once version 11 (YOLOv11) and fire class prediction based on material descriptions using Bidirectional Encoder Representations from Transformers (BERT). In addition, interaction options were generated using a text generation model Fine-tuned Language Net Text-to-Text Transfer Transformer (FLAN-T5). The results indicate that the proposed framework achieves high rendering fidelity and precision, enabling efficient and scalable development of industrial safety training modules.

The Village Credit Institution (LPD) is a microfinance institution that plays a vital role in the rural economy in Bali. LPDs provide credit through a manual and subjective analysis process conducted by loan officers. This often hinders the objectivity and consistency of credit analysis. Modernization is a strategic step to address this issue. This study aims to analyze the digital transformation process in the LPD credit analysis system through the implementation of machine learning. Initial analysis indicates that debtors with long tenors, high delinquency rates, and high debt-to-income ratios have a higher risk of default. This pattern serves as the basis for learning a machine learning model using Logistic Regression, Decision Tree, Random Forest, and XGBoost algorithms to classify creditworthiness. The XGBoost algorithm demonstrated the best performance with an accuracy of 93% and an AUC of 0.96. Regarding credit approval, this model was able to identify high-risk potential debtors with significantly better accuracy than conventional methods. By processing thousands of historical data points in a relatively short time, machine learning accelerates decision-making and significantly improves the efficiency and objectivity of credit analysis. This supports the realization of digital transformation in credit analysis at LPDs.

Driver behavior, vehicle dynamics, and operating conditions strongly influence fuel consumption in open-pit mining operations. This study proposes a driver behavior-aware fuel optimization framework that integrates a digital twin architecture with reinforcement learning to improve fuel efficiency of heavy-duty haul trucks. The framework combines a data-driven vehicle dynamics surrogate, explicit modeling of driver behavior, and proximal policy optimization to enable safe and scalable policy learning within a realistic simulation environment. Historical telematics data were used to construct the digital twin and evaluate the learned policy under controlled operating conditions. Experimental results show that the reinforcement learning agent produces substantially smoother driving behavior, characterized by stable speed regulation and elimination of aggressive acceleration and braking events. Compared to historical operator driving, fuel consumption per kilometer, computed using rollout-level aggregation of cumulative fuel consumption and total traveled distance, was reduced from 4.45 L/km to 3.02 L/km, corresponding to a 32.05% improvement in fuel efficiency. The results demonstrate that explicitly modeling driver behavior within a digital twin-based reinforcement learning framework can yield significant fuel savings while maintaining realistic and interpretable driving strategies. The proposed approach provides a promising foundation for the development of decision-support and driver assistance systems aimed at improving energy efficiency in open-pit haulage operations.

Network Intrusion Detection Systems (NIDS) play a vital role in modern cybersecurity by leveraging artificial intelligence (AI), particularly deep learning (DL) and machine learning (ML), to detect and mitigate malicious activities. However, these AI-driven systems are highly vulnerable to adversarial attacks, where small, imperceptible perturbations in input data can deceive models and significantly reduce detection accuracy. This raises critical concerns about the security and reliability of intrusion detection, especially in real-world scenarios where attackers exploit adversarial transferability to bypass defenses. This research investigates the threat posed by black-box adversarial attacks via surrogate models, focusing on the ability of adversarial examples to transfer across different architectures. This study simulates real-world adversarial threats, demonstrating how attacks crafted on one model can effectively deceive another, compromising NIDS security.  A comparative study is conducted on two widely used AI models: an Artificial Neural Network (ANN) and a Convolutional Neural Network (CNN), both trained on the CICIDS 2019 dataset. The study evaluates the robustness of these models against two gradient-based adversarial attack methods, Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD), to determine their susceptibility under black-box adversarial conditions. Experimental results indicate that CNN-based NIDS are more vulnerable to adversarial attacks than ANN-based models, with adversarial examples successfully transferring across architectures. These findings highlight the critical risks associated with adversarial transferability, underscoring the need for enhanced security measures to strengthen AI-driven intrusion detection systems against evolving cyber threats.

Text document clustering plays a pivotal role in organizing large-scale unstructured data, yet conventional clustering algorithms, such as k-means, often struggle with high-dimensional data, suboptimal initializations, and local minima issues. This paper introduces a novel comparative analysis of three advanced optimization techniques integrated with k-means: Chaotic Northern Goshawk Optimization (CNGO), Mayfly Optimization Algorithm (MOA), and Modified Particle Swarm Optimization (MPSO). This work is unique because it integrates these metaheuristic algorithms to improve clustering performance, targeting initialization challenges and increasing accuracy. Extensive experiments were conducted on benchmark datasets, including Reuters-21578, 20-Newsgroup, and BBC-Sport. All three models outperform traditional k-means in terms of accuracy, Adjusted Rand Index (ARI), Normalized Mutual Information (NMI), and V-measure. This research offers new insights into optimizing clustering processes using metaheuristic algorithms and provides a foundation for future exploration in large-scale document clustering.
Our study is significant because it systematically overcomes key limitations of conventional k-means for high-dimensional text data, such as poor centroid initialization, local minima, and reduced effectiveness on sparse corpora by integrating and comparatively evaluating three advanced metaheuristics (CNGO, MOA, MPSO) with k-means on standard benchmark datasets. The value of this work lies in the consistently improved clustering quality (Accuracy, ARI, NMI, V-measure) achieved by the proposed hybrids, and in showing that MOA–k-means in particular offers a robust, scalable solution for real-world text analytics applications such as information retrieval, recommendation, and topic discovery.