An adaptive learning framework driven by artificial intelligence (AI) was proposed in which cognitive, emotional, and cultural dimensions of learner diversity were jointly modeled to address heterogeneous educational needs in a personalized and inclusive manner. Within the proposed system, learner adaptation was achieved through the coordinated deployment of multiple machine learning paradigms: models based on Decision Trees (DTs) were employed to dynamically align instructional content with learners’ cognitive profiles, Recurrent Neural Networks (RNNs) were utilized to capture temporal patterns in emotional engagement, and Collaborative Filtering (CF) techniques were applied to accommodate cultural preferences. The framework operates as a continuously adaptive system, enabling instructional content to be refined based on learner data derived from a dataset comprising 10,000 students. Experimental evaluation demonstrated that the proposed approach yielded statistically significant improvements in learning outcomes when compared with conventional instructional methods. Specifically, mean quiz and assignment scores were increased by 15.7% and 14.4%, respectively, while emotional engagement indicators exhibited an improvement of 35.8%. In addition, cultural satisfaction metrics were enhanced by 24.2%. These results suggest that the synergistic integration of cognitive, emotional, and cultural adaptation mechanisms contributes substantively to academic performance gains, heightened learner engagement, and improved educational equity. Beyond performance improvements, the proposed framework is designed with scalability and robustness, allowing for deployment across personalized educational contexts. As such, the framework offers a viable pathway for the development of next-generation personalized education systems capable of supporting diverse learners at scale while maintaining pedagogical effectiveness and inclusivity.

Skin burns represent a major clinical concern due to their association with pain, functional impairment, sensory damage, and even life-threatening complications. Early and accurate assessment is critical for first aid, clinical intervention, and the prevention of secondary complications. However, conventional burn diagnosis remains highly dependent on visual inspection and clinical expertise, which can introduce subjectivity and delay timely decision-making. To address these limitations, a hybrid automated skin burn detection framework was proposed, integrating transformer-based feature extraction with classical machine learning classification. In this framework, discriminative visual features were extracted using multiple Vision Transformer (ViT) architectures, including ViT-B/16, ViT-L/16, ViT-B/32, and DINOv2 (a self-supervised Vision Transformer model). The extracted features were subsequently fused. Given the resulting high-dimensional feature space, dimensionality reduction was performed using the Chi-square (Chi$^2$) algorithm, through which 500 features were retained, reducing computational complexity and mitigating the risk of model overfitting. The reduced feature set was then employed for burn classification using six classifiers. Model effectiveness was assessed using accuracy, precision, sensitivity, and F1-score metrics. Experimental results demonstrated that the Support Vector Machine (SVM) classifier achieved the highest classification performance, yielding an accuracy of 82.29%. Comparable yet slightly lower accuracies were observed for the Light Gradient Boosting Machine (LGBM) (80.51%) and Extreme Gradient Boosting (XGBoost) (80.17%) classifiers. Overall, the proposed hybrid model consistently outperformed baseline models, highlighting its superior discriminative capability. These findings indicate that the proposed framework holds strong potential for integration into clinical decision support systems, offering a reliable and objective tool for automated skin burn detection.

Facial expression recognition (FER) remains a challenging problem in computer vision owing to subtle inter-class visual differences, substantial intra-class variability, and severe class imbalance in commonly adopted benchmark datasets. In this study, a statistically rigorous comparative evaluation of two pretrained Convolutional Neural Network (CNN) architectures, MobileNetV2 and EfficientNet-B0, was conducted using the FER2013 dataset. To ensure methodological fairness and reproducibility, both architectures were fine-tuned and evaluated under strictly identical experimental conditions. Model performance was systematically assessed using overall classification accuracy and macro-averaged precision, recall, and F1-score to account for class imbalance, complemented by confusion matrix analysis and multi-class receiver operating characteristic area under the curve (ROC–AUC) evaluation. Beyond conventional performance reporting, the reliability and robustness of the observed differences were examined through McNemar’s test and paired bootstrap confidence intervals (CIs). The experimental results demonstrate that EfficientNet-B0 consistently outperforms MobileNetV2 across all evaluation criteria. Statistical analysis confirms that the observed performance gains are significant at the 5% significance level. These findings provide empirically grounded evidence for informed model selection in FER tasks and highlight the importance of integrating statistical validation into comparative deep learning studies. The results further suggest that EfficientNet-B0 offers a favorable balance between recognition accuracy and computational efficiency, making it a compelling candidate for real-world FER applications, including human–computer interaction, affect-aware systems, and assistive computing environments.

This study introduces a grammar-based, chaotic-oriented programming language, termed ChaosL, to address persistent numerical precision and reproducibility challenges in the computational analysis of chaotic systems. The language, along with its compiler and parser, is designed end-to-end with consideration of chaotic maps. Numerical accuracy is systematically managed through grammar-level precision specification and automated error monitoring mechanisms, enabling exact control over floating-point representations, including single precision, double precision, and arbitrary-precision BigDecimal arithmetic with configurable decimal resolution of up to 100 digits. The proposed grammar natively supports ten widely studied one-dimensional and two-dimensional discrete chaotic maps, which may be composed using newly defined hybrid composition paradigms, namely alternate, blend, cascade, and feedback-driven coupling. To ensure computational reliability, multiple error assessment strategies are integrated, including direct error estimation, shadow computation, and interval arithmetic. In addition, ensemble-based simulation capabilities are incorporated to evaluate trajectory separation and estimate predictability horizons. The automated computation of Lyapunov exponents is embedded at the language level, achieving an accuracy of up to 99.6% while simultaneously enabling code-size reductions of approximately 85–92%. The adaptable architecture of ChaosL establishes a reproducible computational framework for discrete chaos research and facilitates the systematic identification of emergent behaviors in hybrid dynamical systems. Moreover, the design provides a scalable foundation for future extensions toward continuous-time systems, interactive visualization environments, and cloud-based collaborative experimentation, thereby advancing precision-aware computational practices in nonlinear dynamics and chaos theory.

Osteoarthritis (OA) affects approximately 240 million individuals globally. Knee osteoarthritis, a crippling ailment marked by joint stiffness, discomfort, and functional impairment, is particularly the most widespread kind of arthritis among the elderly. To assess the severity of this disease, physical symptoms, medical history, and further joint screening examinations including radiography, Magnetic Resonance Imaging (MRI), and Computed Tomography (CT) scans have frequently been considered. It is difficult to identify early development of this disease as conventional diagnostic methods could be subjective. Therefore, doctors utilize the Kellgren and Lawrence (KL) scale to evaluate the severity of knee OA with visual images obtained from X-ray or MRI. The detection and prediction of the severity of knee OA indeed requires a novel model that uses deep learning models, including Inception and Xception. Utilizing the KL grading scale, the model, including Xception, ResNet-50, and Inception-ResNet-v2 could determine the degree of knee OA suffered by patients. The experimental results revealed that the Xception network achieved the highest classification accuracy of 67%, surpassing ResNet-50 and Inception-ResNet-v2, demonstrating its superior ability to automatically grade OA severity from radiographic images.

Dealing with privacy and security becomes more complicated nowadays with the emergence of big data era. Privacy, data value, and system efficiency should be managed using multiple solutions in current analytics. In this paper, privacy-preserving techniques were selected and reviewed for big data analysis to reduce threats imposed on healthcare data. Various security solutions, including k-anonymity, differential privacy, homomorphic encryption, and secure multi-party computation (SMPC), were programmed and examined using the Medical Information Mart for Intensive Care III (MIMIC-III) healthcare dataset. Assessments were conducted cautiously for each method of data collection in respect of security, time required, capacity of handling large data sets, usefulness of the data, and compliance with regulations. By using differential privacy, it was possible to maintain a balance between privacy and utility by allocating additional resources to the program. The security of data was facilitated by homomorphic encryption though it was not easy to operate and reduce the speed of computer systems. Moreover, achieving scalability in the SMPC required a significant amount of computing power. Although k-anonymity enhanced data utility, it was vulnerable to certain types of attacks. Protecting privacy in big data would limit the performance of systems; for multiple copies of data, scientists can now utilize analytics, differential privacy, and the SMPC, which is highly effective for analyzing private data. However, such approaches should be further optimized to handle real-time processing in big data applications. Experimental evaluation showed that processing 10,000 patient records using differential privacy took an average of 2.3 seconds per query and retained 92% of data utility, while homomorphic encryption required 15.7 seconds per query with 88% utility retention. The SMPC achieved a high degree of privacy with 12.5 seconds per query but slightly reduced scalability. As recommended in this study, the implementation of privacy-focused solutions in big data could help researchers and companies establish appropriate privacy policies in healthcare and other similar areas.

Hyperparameter search was found not making good use of compute resources as surrogate-based optimizers consume extensive memory and demand long set-up time. Meanwhile, projects running with fixed budgets require lean tuning tools. The current study presents Bounding Box Tuner (BBT) and conducts tests of its capability to attain maximum validation accuracy while reducing tuning time and memory use. The project team compared BBT with Random Search, Gaussian Processes for Bayesian Optimization, Tree-Structured Parzen Estimator (TPE), Evolutionary Search and Local Search to decide on the optimum option. Modified National Institute of Standards and Technology (MNIST) classification with a multilayer perceptron (0.11 M weights) and Tiny Vision Transformer (TinyViT) (9.5 M weights) were adopted. Each optimizer was assigned to run 50 trials. During the trial, early pruning stopped a run if validation loss rose for four epochs. All tests applied one NVIDIA GTX 1650 Ti GPU; the key metrics for measurement included best validation accuracy, total search time, and time per trial. As regards the perceptron task, BBT reached 97.88% validation accuracy in 1994 s whereas TPE obtained 97.98% in 2976 s. Concerning TinyViT, BBT achieved 94.92% in 2364 s, and GP-Bayesian gained 94.66% in 2191 s. It was discovered that BBT kept accuracy within 0.1 percentage points of the best competitor and reduced tuning time by one-third. The algorithm renders the surrogate model unnecessary, enforces constraints by design and exposes solely three user parameters. Supported by the evidence of these benefits, BBT was considered to be a practical option for rapid and resource-aware hyperparameter optimization in deep-learning pipelines.