Handwriting analysis has proven to be a valuable tool in diagnosing neurological diseases and neurodevelopmental disorders. Among these, Attention Deficit Hyperactivity Disorder (ADHD) stands out as a significant focus. ADHD affects how the brain manages attention, impulsivity, and hyperactivity. In this context, this study proposes a novel method to ADHD prediction in children by integrating handwriting-based features within a hybrid learning analysis. Specifically, we focus on combining the strengths of multiple base models in a voting classification framework to achieve improved accuracy and robustness. Such a proposed hybrid learning approach is compared with single learning as well as ensemble learning approaches. In addition, we focus on feature engineering that explores handwriting patterns and dynamics to identify the most informative characteristics for accurate ADHD detection. The experimental results demonstrate the effectiveness of the proposed method, showing promising accuracy in distinguishing children with ADHD from healthy subjects.
In this paper, we propose a robust and blind image watermarking scheme based on Quantization Index Modulation (QIM), utilizing Discrete Wavelet Transform (DWT) and Discrete Cosine Transform (DCT). The method adaptively adjusts the watermark bit strength with proper pixel value modification, embedding it into the middle-frequency band of DCT coefficients and selected DWT subbands. The embedding strength is dynamically tuned based on local coefficient magnitudes, ensuring minimal visual distortion while maintaining high resistance to common image processing attacks-particularly filtering, compression, and noise. To improve security and unpredictability-unlike traditional random bit generators with fixed seeds-our proposal uses a Physical Unclonable Function (PUF) device, which provides a hardware-rooted source of entropy that is extremely difficult for attackers to predict. This approach dynamically inserts watermark bits, satisfying all core properties of a robust watermarking benchmark. The proposed method is fully blind, and the experimental results show an outstanding Peak Signal-to-Noise Ratio (PSNR $\gt80 \mathrm{~dB}$), making our scheme a powerful method in terms of robustness and imperceptibility. Additionally, the Structural Similarity Index (SSIM) reaches near-perfect levels, and the Bit Correction Rate $(B C R=100 \%)$ confirms the method’s efficiency and visual fidelity.
Intrusion Detection Systems (IDS) form a critical component of the security infrastructure in Software-Defined Networks (SDN). However, they often need to strike a balance between high detection accuracy, real-time responsiveness, and adaptability. Furthermore, traditional IDS that operate at the controller level suffer from centralized bottlenecks and high false alarms for evolving threats. Thus, edge-assisted approaches have been proposed to alleviate some of these drawbacks. While emerging quantum-enhanced strategies can be employed to accelerate this process, this would require specialized hardware at the edge devices, which is either impractical or too cost-prohibitive. In this work, we apply Classical Client Quantum Federated Learning (CC-QFL) to adaptive resource-constrained SDN nodes that use classical shadow tomography to offload quantum model training and aggregation to a centralized quantum controller. The overhead of periodic retraining is significantly reduced and inference requires $0.1 \mu$ s per instance. CC-QFL is evaluated on the Edge IIoTset dataset consisting of 1,048,576 instances, achieving $83.2 \%$ accuracy, $86 \%$ precision, $95 \%$ recall, and a $91 \%$ F1-score. It takes only 2.8 ms per round to train, which is several orders of magnitude faster than conventional deep learning IDS. These results reveal that in dynamic SDN environments, CC-QFL provides ultra-low latency, scalable, and robust intrusion detection, opening up possibilities for quantum-driven enhancements in the responsiveness of SDN controllers.
Patients with Cerebral Palsy (CP) often require specialized, proactive care to avoid preventable emergency department visits. Traditional healthcare models are reactive, responding to events after they occur rather than predicting and preventing them. This paper presents a novel machine learning framework that leverages patient health records to predict four critical health outcomes for CP patients: emergency department visits, medication needs, diagnostic trends, and procedure requirements. Our approach utilizes deep learning models trained on synthetic FHIR (Fast Healthcare Interoperability Resources) data from over $\mathbf{4, 0 0 0 ~ C P}$ patients to create comprehensive patient timelines with rich clinical features. The emergency visit prediction model achieves $97.5 \%$ AUC on test data, demonstrating strong discrimination between high and lowrisk patients. The multi-label prediction models show varying performance based on dataset size and label frequency, with Top20 models significantly outperforming Top-50 models: medication prediction achieves F1-scores of 0.78 (Top-20) versus 0.19 (Top50), while diagnosis and procedure prediction models achieve 0.72 versus 0.018 (Top-50) and 0.75 versus 0.22 (Top-50) F1scores, respectively. Feature importance analysis reveals that recent healthcare utilization patterns, medication regimens, and diagnostic history are the strongest predictors of emergency care needs. We also present a novel timeline-based feature engineering methodology that captures temporal trends in patient health. Ablation studies demonstrate that the inclusion of derived features, such as healthcare utilization metrics and medication complexity indicators, significantly improves model performance. The framework can assist healthcare providers in identifying high-risk patients who would benefit from preventive interventions, potentially reducing emergency visits, improving quality of care, and decreasing healthcare costs. The system’s architecture accommodates continuous learning with new patient data, making it adaptable to evolving clinical patterns and individual patient histories.
The dynamic cybersecurity threats in specialized domains like space systems create significant challenges for educational content development. Traditional curriculum development struggles to keep pace with dynamic industry requirements, taking months to develop and requiring large expert teams. This paper introduces a systematic framework that integrates Generative AI (GenAI) with established instructional design principles to rapidly develop domain-specific cybersecurity training. Our framework combines Retrieval Augmented Generation (RAG) with the Analysis, Design, Development, Implementation, and Evaluation (ADDIE) model, enabling requirements-driven curriculum development that translates industry stakeholder interviews and job descriptions directly into comprehensive educational materials. We demonstrate framework feasibility through Space Information Systems Security Officer (ISSO) curriculum development, generating 500+ domainspecific Knowledge, Skills, and Tasks (KSTs), modular lectures, 886 assessment questions, and gamified exercises within hours rather than months. Market research with 33 industry professionals revealed critical gaps in existing training frameworks, with $82 \%$ emphasizing soft skills and $67 \%$ requiring holistic system understanding beyond traditional cybersecurity domains. Our proof-of-concept implementation with 5 completing participants showed promising results with test scores improving from $73.9 \%$ to $92.1 \%$, though larger validation studies are needed. The systematic framework addresses identified gaps in current cybersecurity training approaches while providing a replicable methodology for other rapidly evolving technical domains requiring specialized workforce development.
