Abstract The rapid growth of IoT networks demands privacy-preserving intrusion detection systems that operate on distributed data without exposing sensitive traffic information. This paper presents BlockFedIDS, a blockchain-enabled federated learning framework that eliminates the centralised aggregator using smart contracts. The framework integrates Differentially Private Stochastic Gradient Descent (DP-SGD) for differential privacy, Multi-Krum for Byzantine-resilient aggregation, and Pedersen commitments for secure model update verification. Experimental evaluation on the NSL-KDD and CIC-IDS2017 datasets with 20 IoT clients achieves 97.14% detection accuracy and 0.984 F1-score, while reducing poisoning-induced performance degradation by 74.3% compared with FedAvg and requiring only 1.4 KB of on-chain storage per training round.
Abstract Although SDN provides a programmable, centrally managed framework for modern networks, that same centralization leaves it exposed to attacks such as Distributed Denial of Service (DDoS). This paper proposes an intrusion detection framework that couples Z-Isomorphic Sigmoid Graph Neural Networks (ZIS-GNN) with Bonobo-Optimization-based (EKPC-BOA) feature selection. The sigmoid-based activation strengthens the graph representation relative to conventional GNNs, capturing complex traffic patterns more faithfully, while the hybrid selector – combining the Bonobo Optimization Algorithm with an entropy score and Pearson correlation – distils the most informative features from the traffic data and thereby improves both efficiency and accuracy. Experiments demonstrate that the proposed ZIS-GNN+EKPC-BOA model attains an accuracy of 97.36%, a precision of 97.37%, and an F1-score of 97.58%, outperforming baseline models such as DNN (89.60%), LSTM (91.68%), BiLSTM (93.77%), and GNN (95.86%), as well as the standard graph baselines GCN (96.18%) and the attention-based GAT (96.58%). The results show the effectiveness of combining graph-based learning with hybrid feature selection for intrusion detection in SDN.
Abstract The rapid growth of social media requires privacy-preserving approaches for detecting toxic, fraudulent, misleading, and inappropriate content while protecting sensitive user information. This paper proposes a trust-aware federated learning framework that enables collaborative model training without sharing raw user data. An attention-based Long Short-Term Memory (LSTM) network captures important semantic relationships in textual content, while a trust scoring mechanism identifies unreliable or potentially malicious clients. A blockchain layer provides secure verification of model updates and enhances transparency throughout the federated training process. Experimental results demonstrate strong classification performance with limited blockchain overhead, highlighting the framework’s potential as a scalable, secure, and efficient solution for privacy-preserving social media content analysis.
Abstract The rapid digitization of healthcare has brought Electronic Health Records (EHRs) to the forefront of clinical data management; however, persistent challenges of centralized control, privacy breaches, absence of patient data ownership, and the inability to support decentralized scientific collaboration continue to impede scalable healthcare research ecosystems. Recent advances in Decentralized Science (DeSci) introduce a paradigm shift by leveraging blockchain, cryptographic primitives, and decentralized governance to enable transparent, trust-minimized, and collaborative biomedical research. This paper proposes a DeSci-driven lightweight hybrid blockchain framework designed to support privacy-preserving and incentive-aware decentralized healthcare research infrastructure. The framework integrates a permissioned blockchain with a lightweight hybrid PBFT–PoA consensus protocol, off-chain storage, and Zero-Knowledge Proof (ZKP)-based authentication to enable secure, privacy preserving data access without disclosing user identity. A tokenomics-based DAO governance layer is incorporated to support decentralized engagement, transparent policy enforcement, and incentive-driven research participation. The proposed system is evaluated through simulation under varying network conditions, with key performance metrics — latency, throughput, and computational cost — assessed across network sizes from 10 to 50 nodes. Simulation-based projections suggest that the proposed framework may achieve lower latency, higher throughput, and improved computational efficiency relative to literature-reported values for MedRec, FHIRChain, and HealthChain under the stated modeling assumptions; these comparisons are model-based and illustrative rather than measurements obtained from a controlled, identical-environment deployment. Beyond data management, the framework enables a DeSci-oriented research lifecycle encompassing decentralized data contribution, validation, and provenance tracking. The simulation-only nature of the current evaluation is explicitly acknowledged as a limitation, with a clear roadmap toward prototype-level implementation on Hyperledger Fabric or Ethereum as immediate future work.
Abstract This paper presents a secure blockchain-enabled Automatic Number Plate Recognition (ANPR) framework for Intelligent Transportation Systems (ITS), combining lightweight YOLO-based license plate detection, OCR-based character recognition, and Ethereum smart contracts for tamper-resistant vehicle data management. The study investigates the inference performance of recent object detection architectures, namely YOLOv11, YOLOv12, and YOLO26, on Raspberry Pi 5 and NVIDIA Jetson Nano edge devices. Furthermore, optimizations for ONNX and TensorRT are explored to enhance inference efficiency on embedded hardware. Experimental results demonstrate complementary strengths among the evaluated models. YOLOv11n achieved the best overall detection performance, obtaining the highest recall (0.945), F1-score (0.9645), and mAP@0.5:0.95 (0.692), while YOLOv11n and YOLOv12n achieved identical precision (0.985) and mAP@0.5 (0.963). YOLO26 demonstrates the best deployment efficiency, reducing inference latency to 33.4 ms on Raspberry Pi (ONNX) and 38.9 ms on Jetson Nano, making it suitable for edge applications. Model optimization significantly accelerates inference, reducing YOLOv11 latency on Raspberry Pi from 98.2 ms to 38.3 ms after ONNX conversion. The results demonstrate that the proposed framework effectively balances detection accuracy, computational efficiency, and data security, making it a promising solution for smart parking and next-generation ITS applications.
