With the advent of fast-growing Internet of Things (IoT) technologies, intelligent creative systems offer new opportunities for adaptive digital media output. Current computational art design frameworks, however, do not sufficiently incorporate IoT-driven human-machine interaction mechanisms with deep-learning-based aesthetic and style intelligence. Accordingly, this paper introduces IoT ArtNet, a novel IoT-based intelligent multimodal framework for adaptive digital media creation. To achieve three different deep learning tasks, IoT ArtNet uses the Art Style Representation Module trained on the WikiArt Art Movements and Styles dataset for artistic style classification, the Aesthetic Intelligence Module trained on the AVA (Aesthetic Visual Analysis) dataset for aesthetic quality prediction, and the Gesture-Based Human-Machine Interaction Module trained on the LeapGestRecog dataset for gesture command recognition using sensors. The framework achieves multimodal fusion of style embeddings, aesthetic quality scores, and gesture control via attention-weighted fusion, enabling adaptive media output decisions in a new Adaptive Media Fusion Module. The framework includes edge and cloud inference layers to support low-latency, IoT-oriented deployment. Offline prototype benchmark evaluations indicate that IoT ArtNet achieves 93.2
The rapid growth of IoT networks has increased their exposure to Distributed Denial-of-Service (DDoS) attacks, particularly HTTP flood attacks that can disrupt service availability in resource-constrained environments. To address the challenges of high-dimensional flow-based traffic data and the need for accurate and efficient detection, this study proposes a hybrid PSO–GWO feature selection framework for HTTP flood-based IoT DDoS attack detection. Particle Swarm Optimization (PSO) and Grey Wolf Optimizer (GWO) are applied independently, and their selected subsets are combined after optimization using post-selection union and intersection strategies. Their outputs are then combined using two feature construction strategies: union (PSO∪GWO) and intersection (PSO∩GWO). The union strategy preserves broader complementary feature information, while the intersection strategy produces a compact subset for lightweight screening. Experiments are conducted specifically on the IoT-DDoS-HTTP_Flood subset of the CIC IoT-DIAD 2024 dataset; therefore, the reported findings are limited to this attack scenario. Among the evaluated classifiers, XGBoost achieved the highest mean performance. Results show that PSO∪GWO with XGBoost reaches 99.38
Edge-IoT intrusion detection systems must balance predictive performance with the throughput, latency, and resource constraints of embedded deployment. This study investigates when hardware acceleration is beneficial for tabular IoT intrusion detection on a Raspberry Pi 5 equipped with a Hailo-8 AI accelerator. We evaluate lightweight multilayer perceptrons, conventional 1D convolutional neural networks, and a quantization-friendly 1D-CNN across CICIoT2023 and Edge-IIoTset using ONNX Runtime, Hailo quantized-emulator evaluation, and physical HailoRT benchmarking. The results reveal an execution-regime-dependent acceleration boundary. For the evaluated MLPs, CPU throughput exceeds HailoRT streaming throughput under both reported CPU reference regimes, whereas the conventional 1D-CNNs achieve higher HailoRT streaming throughput than both CPU references. Relative to CPU batch-size-one execution, the CICIoT2023 and Edge-IIoTset 1D-CNNs reach streaming-throughput ratios of 10.34 × and 9.61 × , respectively; the corresponding ratios relative to buffered CPU execution are 1.88 × and 1.82 × . The quantization-friendly model reduces ONNX-to-quantized-emulator macro-F1 degradation by 7.66 percentage points relative to the standard Edge-IIoTset 1D-CNN while also reducing parameter count, HEF on-disk size, and accelerator-side hardware latency. These findings show that accelerator selection for Edge-IoT IDS should consider model structure, execution regime, runtime overhead, and quantization behavior jointly.
Adult education frequently encounters complex socioeconomic and motivational barriers; predicting continuing education dropouts is a significant challenge. This research proposes a Bayesian network modeling approach to efficiently predict the likelihood that at-risk students will drop out and to intervene before it’s too late. Financial stress, course involvement, and personal drive are influencing variables, but standard machine learning models do not capture the uncertainty of causal links or make them interpretable. To address these shortcomings, the paper presented a Bayesian Network-Based Probabilistic Inference Framework (BN-PIF) that enables real-time probabilistic reasoning by modeling the conditional relationships among pupils’ attributes. Learner data from the past is used to train the framework to identify high-risk students and predict who is likely to drop out. It powers a decision-support system that, based on each student’s risk profile, proposes individualized interventions like academic counselling or financial aid. The proposed BN-PIF is both more predictable and easier for decision-makers to understand. Instead of claiming a 15
The convergence of robotics and the Internet of Things (IoT) has transformed multiple sectors through autonomous, interconnected systems. Despite significant scholarly output, the knowledge structure of Robotics-IoT research remains fragmented. This study provides an integrated scientometric and bibliometric mapping of global Robotics–IoT/IoRT research, combining performance analysis, collaboration-network mapping, thematic mapping, and LDA-based topic modelling to reveal the field’s intellectual structure, collaboration architecture, dominant clusters, emerging themes, and future research frontiers. A scientometric analysis was conducted on 21,733 unique documents retrieved from Scopus and Web of Science after screening and deduplication of 23,377 initial records, covering 1993–2025 across 2,284 sources and over 44,000 indexed author-name records/clusters, cleaned via the PRISMA protocol. Biblioshiny and VOSviewer were employed for publication trends, citation structures, and collaborative networks, while LDA was applied for topic modelling. Robotics-IoT research shows rapid growth post-2016. Using all-author affiliation-country document-level full counting, India, China, and the USA lead publication output. Among the ten most productive countries, Australia shows the highest average citations per document, followed by Canada, the USA, and the United Kingdom. Collaboration networks reflect extensive global partnerships, with emerging economies demonstrating substantial international participation. Thematic analysis identifies major research streams related to industrial digitalization, IoT–robotic systems, energy-efficient networking, machine-learning-based analytics, robotics and sensing, cybersecurity, and edge/cloud computing, alongside specialized applications in agriculture and healthcare. Observed distributions show varying degrees of descriptive alignment with classical scientometric laws (Lotka’s, Bradford’s, Price’s, Zipf’s, Matthew Effect, and Pareto Principle). The originality of this study lies in the domain-specific application and empirical synthesis of these approaches to the global Robotics–IoT/IoRT research landscape. Specifically, the study provides a longitudinal analysis covering 1993–2025, examines the intellectual, social, publication and thematic structures of the Robotics–IoT/IoRT domain, evaluates a comparatively broad set of classical scientometric regularities, and complements keyword-based analysis with abstract-based LDA topic modelling. The study consequently contributes new empirical evidence concerning the evolution, concentration, collaboration, intellectual structure and thematic development of Robotics–IoT/IoRT research.
