The Internet of Energy is a critical national infrastructure that must be protected against security attacks to prevent significant economic losses. Its communication infrastructure integrates electrical power grids with data analytics tools to automate energy generation, transmission, distribution, and consumer-related operations. However, it remains vulnerable to cyber-physical attacks and sensitive data leakage. Recent federated split learning-based intrusion detection systems help preserve data privacy but face significant challenges in large-scale heterogeneous environments. The literature highlights the need to address data, computational, and communication heterogeneity, which can reduce model performance. Therefore, this prospective paper proposes a hierarchical federated split learning framework that distributes processing across local, regional, and global levels while considering cross-layer dependencies. Graph neural networks are integrated to model network topology and provide structured feature embeddings for multi-agent deep reinforcement learning. This integration aims to understand complex infrastructure dependencies, optimize conflicting parameters, coordinate actions, and automate responses. The proposed framework aims to shift cyber-physical defense from centralized, reactive detection towards collaborative, proactive, and predictive systems. It also aims to enhance situational awareness, risk assessment, continuous adaptation, threat mitigation, and response optimization to support the early detection of complex and unknown cyber-physical threats, including coordinated attacks across critical infrastructure, thereby promoting a proactive, cognitive, and sustainable defense ecosystem.
Military enlistment decisions are shaped by complex, multidimensional factors, including socioeconomic status and personal attitudes. This study develops a predictive decision model to identify the key factors that distinguish voluntary enlistment from conscription by lottery in Thailand's 23rd Military District. A dataset of 200 military personnel (147 volunteers and 53 drafted personnel) was analyzed using data mining techniques in Weka. The Random Tree and Random Forest algorithms were employed alongside three baseline classifiers: logistic regression, Naive Bayes, and J48. The models were evaluated based on accuracy, precision, recall, F1 score, and kappa statistics. Random Forest achieved the best balance between overall accuracy and sensitivity to the minority class (Accuracy = 92.86%, Kappa = 0.814, Weighted F1 = 0.926, Lottery-class F1 = 0.861; 10-fold cross-validation accuracy = 95.95% ± 0.44), outperforming all baseline models, particularly in classifying the minority (lottery/drafted) class, for which logistic regression achieved an F1 score of only 0.706. The most influential demographic predictors were occupation, province, and income. When attitudinal variables were included, cultural factors became the most important factor, followed by risk and legal perceptions. However, predictive accuracy did not improve compared with using demographic variables alone. These findings show that machine learning can effectively model enlistment behavior. The Royal Thai Army should develop recruitment policies that target specific occupations and geographical areas while strengthening cultural messaging and risk communication to increase voluntary enlistment rates.
Breast cancer is a major health problem worldwide, particularly in Saudi Arabia, where it is often diagnosed in younger women and at more advanced stages than in Western countries. Breast self-examination (BSE) is a low-cost method, but knowledge and practice remain limited. This study aimed to assess the effect of a nursing-led BSE education program on the knowledge and practice of non-medical female staff. A one-group pretest–posttest study was conducted with 60 female staff at Jouf University. The program was based on Roy’s Adaptation Model and included structured teaching sessions and practical demonstrations using a silicone breast model. Participants were assessed before the program, immediately afterward, and two months later. A validated knowledge test and a 20-item BSE checklist were used. Mean knowledge scores increased from 5.17 (SD = 1.74) at baseline to 9.18 (SD = 0.43) after the intervention. Mean practice scores increased from 1.53 (SD = 2.46) to 20.0 (SD = 5.6). Improvements remained at follow-up. Both outcomes showed statistical significance (p = 0.001). The findings suggest that nursing-led education may improve BSE knowledge and practical skills among female university staff.
This study investigated the effectiveness of Brain-Based Learning (BBL) in improving Grade 7 students’ academic achievement in the topic of exponents. A one-group pretest–posttest design was employed with 23 students over five weeks. The findings indicated that the BBL instructional plan exceeded the 75/75 effectiveness criterion and resulted in significant improvements in posttest scores. Students reported very high satisfaction with the learning process. Qualitative observations also confirmed active engagement and conceptual understanding. The study demonstrates that BBL supports both the cognitive and emotional aspects of learning, making it a promising approach for teaching abstract mathematical concepts in lower secondary education.
The role of coronary angiography is important for visualizing coronary arteries and detecting abnormalities such as coronary artery disease (CAD). Nevertheless, its diagnostic value is often constrained by image noise, poor contrast resolution, motion artifacts, and changes in vessel curvature and orientation. To overcome these challenges, this study proposes an automated CAD segmentation and diagnosis system using angiographic X-ray images (invasive coronary angiograms). The proposed framework begins with an image preprocessing step using two hybrid noise reduction and contrast enhancement algorithms, namely, the Median Filter with Optimized Maximum Contrast and the Adaptive Median Filter with Optimized Maximum Contrast, to reduce noise without losing edge information. For coronary artery segmentation, a Spider U-Net is adopted, which complements the traditional U-Net with attention mechanisms to enhance feature saliency and focus on clinically significant regions. The proposed framework is evaluated using several statistical measures, with accuracy, precision, recall, F1-score, and Jaccard similarity coefficient values of 0.9549, 0.8510, 0.7836, 0.9008, and 0.8762, respectively. The experimental results demonstrate that the proposed framework outperforms existing methods by effectively reducing background artifacts and enhancing the clarity and precision of vascular structures, thereby improving the reliability of coronary artery disease diagnosis.
