This paper describes the design, development, and testing of a high-speed, economical, and efficient wireless text data transmission system through Visible Light Communication (VLC), also known as Light Fidelity, for underground metal mines. It deals with the importance of a stable communication system for underground situations, where Radio Frequency communication systems are limited. The proposed work consists of the design, fabrication, and analysis of a Light Fidelity communication system setup, coupled with Arduino microcontrollers. The simulation test is performed through Proteus ISIS software for the validation of circuit designs. Voltage versus Distance analysis demonstrates the system’s effectiveness, as the data transmission is observed without distortion up to a certain voltage threshold. The study concludes that Li-Fi may offer a promising solution for enhancing communication and resolving line of sight problem in underground mines, using existing LED (Light Emitting Diode) lighting infrastructure for transmission. In this study, the analysis of the distance vs. voltage depicted that when a minimum 25% of the input voltage from the transmitter end is received at the receiver’s end, then the data transfer occurs without any distortion.
Material Requirement Planning (MRP) is an important part of production management because it ensures the timely availability of materials while reducing costs and surplus inventory. This study introduces a Goal Programming (GP)-based optimization model for material planning in a furniture manufacturing company that produces three product types: dining tables, folding chairs, and fittings. The suggested methodology combines MRP concepts with preemptive multi-objective optimization to address common challenges in the manufacturing process and ultimately to reduce production costs, inventory holding costs, and expenses related to resource idle time and overtime. The mathematical formulation is built on a preemptive priority structure and solved using two computing approaches: Microsoft Excel Solver for baseline linear programming cost minimization and MATLAB's goal attain function for multi-objective goal programming. The suggested model achieves zero excess in priority production goals while keeping total production costs at 10.3% of the LP-derived optimum. The combination of MRP and Goal Programming is demonstrated to provide a practical, scalable, and computationally efficient decision-support framework for production managers in small and medium-sized manufacturing industries, outperforming conventional single-objective planning approaches reported in the literature.
In recent years, the integration of telemedicine has been transforming different environments. In the context of health, the incorporation of contactless monitoring systems supported by telemedicine has become a relevant option in the way of monitoring patients remotely and facilitating access to health services, however, its literature is dispersed. Therefore, this study aimed to conduct a systematic review that encompasses the current evidence of telemedicine as a tool to accurately monitor patients remotely in real time. Applying the PRISMA methodology, 55 relevant documents were extracted from the Scopus, WoS, Pubmed and IEEE Xplore databases, among them 24 keywords were identified, of which "telemedicine" and "patient monitoring" stand out. In addition, it was identified that the United States stands out in co-authorship between countries, and since 2016, scientific production has been gradually increasing. Content-specific results indicated that 49 papers directly related to the impact of telemedicine technologies, 3 focused on opportunities for optimizing the quality of remote care and predictive analytics, and the remaining 3 explored the limitations and challenges of contactless monitoring systems. The review concludes by stating that this type of technology has great potential to support health professionals, however, it is necessary to improve and address the limitations.
Rare diseases affect approximately 300 to 400 million individuals around the world, posing major difficulties in medical diagnosis. Among these, rare skin disorders represent a significant subset, frequently characterized by intricate visual patterns and higher inter-class similarity, which creates substantial obstacles in precise diagnosis. Artificial Intelligence (AI)-driven research for rare skin diseases has accelerated rapidly, unlocking new potential for timely, precise diagnosis and better long-term management strategies. DL techniques have considerably enhanced the classification and detection of skin diseases, including rare skin diseases, across clinical image analysis. However, the availability of adequate labeled datasets for rare skin diseases is limited, which restricts the generality of deep learning systems. To address these challenges, this study presents a Progression-Aware Synthetic Learning Framework for Rare Skin Disease Diagnosis (PASLF-RSDD). The primary objective of this study is to enhance rare skin disease diagnosis through multi-modal synthetic data augmentation, thereby improving the performance of the Deep Learning Model. Initially, the proposed PASLF-RSDD model employs Multi-Modal CycleGAN for synthetic image generation. Following that, high-level discriminative features are extracted using DenseNet121 integrated with Squeeze-and-Excitation blocks for improved feature representation. The extracted features are then passed into a bidirectional convolutional long-short term memory network for accurate rare skin condition classification, such as Elastosis Perforans Serpiginosa, Lentigo Maligna, Nevus Sebaceus, and Blue Naevus. The hyperparameters of the proposed BiConvLSTM classifier are automatically optimized by the Ant Lion Optimizer (ALO), which can effectively search the hyperparameter space to find optimal parameter combinations. This optimization approach stabilizes convergence, reduces the manual selection of parameters in the optimization, and improves the classification performance. Eigen-CAM is embedded to produce explanations in the form of images that make the model more interpretable and boost confidence in the diagnosis predictions by clinicians. The proposed PASLF-RSDD framework was evaluated experimentally with a benchmark dataset called DermaEvolve. Comparative analysis shows that they have better classification results in several evaluation metrics.
The Industrial Internet of Things (IIoT) has expanded the attack surface of modern industrial systems, creating a greater need for tamper-evident, auditable, and low-latency security mechanisms. We reviewed 44 peer-reviewed journal articles published between 2019 and 2024. Under the PRISMA 2020 framework, these studies were selected from 26,994 records retrieved from Scopus, ScienceDirect, IEEE Xplore, and the ACM Digital Library. For each included paper, we recorded the industrial context, security objective, role of blockchain, complementary technologies, validation approach, and reported limitations. The studies were subsequently organized into seven primary themes for descriptive synthesis: foundations and reviews; integrity and privacy; architecture and edge integration; 5G/6G convergence; authentication and access control; AI-enabled intelligent security; and trust, traceability, and industrial applications. Across the corpus, blockchain is used most often for device authentication, data integrity, access control, traceability, secure data exchange, and distributed trust management. Permissioned and edge-assisted architectures are generally regarded as more suitable for industrial constraints than resource-intensive public-chain designs. Even so, scalability, interoperability, latency, energy consumption, smart-contract robustness, and limited long-term industrial validation remain unresolved. Publications appearing after 2024 are discussed separately as a contextual update on Edge AI, TinyML, mobile edge computing, and zero-trust integration; they are not included in the 44-study review corpus.
The growing interest in multilevel inverters for high-power applications is largely attributable to their ability to lower Total Harmonic Distortion (THD) in the output voltage and to the reduced blocking voltage requirement for the switching devices. Presently, these inverters consist of series configurations of fundamental building blocks, each supplied by dedicated, constant DC voltage sources. The inverters have been examined in both symmetric and asymmetric operation modes to generate an expanded palette of voltage levels. The present study extends this architecture by replacing the conventional DC voltage sources with photovoltaic (PV) cells, whose output voltages vary with solar irradiation. The work of Takahashi and Yoshiharu (2002) provides the guiding framework for employing Maximum Power Point Tracking (MPPT) to extract optimal voltage from the PV source. Given that the input to the multilevel inverter should ideally remain constant, a flyback forward converter is interposed between the PV array and the inverter to stabilize the input voltage while accommodating the non-constant solar output. The flyback converter delivers a set of multiple constant output voltages, which the multilevel inverter then expands to a total of 49 discrete voltage levels. The combined system is analyzed to quantify both the total power dissipation within the switching devices and the Peak Inverse Voltage (PIV) experienced. The theoretical predictions are validated through a comprehensive simulation conducted in MATLAB. Results have been confirmed via experiments using bench-scale prototype systems.
This paper reports the design, implementation and deployment of an open-source Learning Management System (LMS) architecture supporting hybrid pre-laboratory instruction in a resource-constrained university setting. The platform was deployed with Tutor, the Docker-based distribution of Open edX, on a single virtual private server with four virtual CPU cores and 4 GB of RAM, with DNS and TLS resolution provided by Cloudflare. The architecture is layered across infrastructure, platform, course content and data processing. Two pre-laboratory modules were delivered through short instructional videos, auto-graded formative quizzes and an embedded interactive serious game, and the engineering trade-offs of third-party tool integration without a Learning Tools Interoperability bridge are analysed. A four-stage learning analytics pipeline extracts learner activity from the LMS gradebook, selects valid engagement signals and computes a composite engagement index. Operational validation over a continuous 58-day production period shows a mean edge response time of 67.3 ms (SD 14.0 ms, n = 20), an application-tier saturation throughput of approximately 74 requests per second, and zero failed requests across 1,600 load-test requests at concurrency levels from 5 to 50. Across the deployed cohort, 82.4% of learners in Module CS101 (n = 17) and 77.3% in Module CS102 (n = 22) attempted at least one formative quiz, and the composite engagement index (n = 17) reached a mean of 76.5% (SD 33.8%, median 88.9%). The result is a low-cost, reproducible blueprint for hybrid digital instruction, together with a quantified account of its capacity envelope and its limits.
Environmental variability, pest infestation, and resource inefficiency have become increasingly problematic to agriculture, and on-line decision making using intelligent, adaptive technologies has been called for. This research is suggesting a novel Integrated Artificial Intelligence Decision Support System (AI-DSS) in precision agriculture so as to attain proper context-aware analysis and recommendation across dynamic field conditions. The framework is based on lightweight deep learning frameworks (MobileNetV2, Efficient Net-lite) and a newly designed Adaptive Feature Optimization (AFO) engine, which dynamically reweights convolutional features based on the temporal stability and environment consistency. Mathematically, the AFO mechanism is obtained via a weighted pooling of the instantaneous features of CNN with the temporal averaged prototypes using adaptive attention weights, which filter out transient noises due to illumination variations, occlusion, and sensor noises. The optimized features are fused with data from environmental and soil sensors in a hybrid Decision Support System (DSS) based on rule-based reasoning, Bayesian inference, and temporal tracking to construct explainable and region-specific recommendations to farmers. Experimental evaluations using the experimental data PlantVillage, DeepWeeds, and Fieldstream Sim demonstrate the superiority of the proposed AFO enhanced framework over the baseline CNN classifier in terms of classification accuracy (.95), macro F1 score (.95), and robustness to distortions (.15% improvement). Additionally, when deployed on edge devices like Raspberry Pi 4 and Nvidia Jetson Nano, the system achieves real-time inference latency (<200 ms) and low energy consumption (~520 mJ/frame), which validates the scalability of the system in low-resource settings. In addition, the expert agreement was enhanced with the DSS module integration to 91.4% with 57% less false alarms. The obtained results validate the proposed AI-DSS framework with AFO as a promising solution to cover the distance between accuracy in a controlled lab environment and reliability in the field, providing a strong, explainable, and resource-efficient solution to the digital sustainable agriculture problem. This solution is further being expanded into proactive farm intelligence and climate resilience through multimodal sensing, satellite assisted crops monitoring, and adaptive decision-making-led solutions.
Engineering ballistic materials development, optimization and improvement are primarily backed by scientific, experimental and mathematical theories and findings. This has led to the development of numerical strategies that are widely considered to effectively yield cost-effective and rapid materials development, and optimization for continuous improvement of existing materials. These numerical strategies incorporate the structural model, computations and analysis of materials data and have been intensively being applied in metallic ballistic materials. However, in ballistic polymer materials such as Ultra-High Molecular Weight Polyethylene (UHMWP-E), the numerical strategies are still at a developing stage. This paper attempts to provide a comprehensive review of constitutive models and numerical simulation technique models applied in ballistic materials for guidance in obtaining improved results and future use of projectile penetration models in UHMWP-E polymers.
Modern healthcare systems are experiencing severe problems in both forecasting comorbid conditions and patient privacy amid decentralized healthcare facilities. This paper introduces a single framework in which multi-task deep learning and secure multi-party computation are combined to result in privacy-preserving disease prediction and risk assessment. The suggested Multi-Task Modified Deep Neural Network (MT-MDLNN) system with five conditions proposed simultaneously to determine cardiovascular disease, diabetes mellitus, chronic kidney disease, hypertension, respiratory disorders and bi- or inter-disease associations. A new Correlation Aware Hybrid Whale-Coati Optimization (CA-HWCO) algorithm explicitly captures comorbidity patterns and it attains better convergence and clinically significant disease relationship learning. To achieve collaborative training between healthcare institutions without sharing sensitive patient information, the framework uses secure multi-party computation protocols that are based on Modified Elliptic Curve Diffie-Hellman (M-ECDH) cryptography as well as secret sharing schemes. It is experimentally validated on six publicly available healthcare datasets with extraordinarily high mean classification accuracy (92.4) and can recognize pattern of clinically validated comorbidity such as diabetes-kidney disease correlation (0.76), as well as cardiovascular-hypertension patterns (0.79). Secure aggregation protocols add as little as 6-9% overhead to communication but information theorems not only in order distance but also in image value fidelity. Risk stratification analysis depicts that 89.4% are in accord with expert clinical examination with real-time inference ability of 38ms per patient. This study provides an overall remedy towards privacy-sensitive collaborative healthcare intelligence by allowing medical institutions to come up with precise multi-disease prediction models without failure to the privacy of patients.
The rapid growth of online learning platforms has generated large volumes of educational data that can support student performance prediction. However, existing Machine Learning (ML) and Deep Learning (DL) approaches often face challenges related to generalizability, feature integration, and early identification of at-risk students. This study presents a comparative evaluation of classical ML, boosting-based, DL, and a proposed Hybrid Artificial Neural Network-Deep Fusion Model (ANN-DFM) using the Open University Learning Analytics Dataset (OULAD). A unified experimental framework incorporating data cleaning, feature selection, and class balancing was employed to ensure fair benchmarking. Results show that Logistic Regression and Support Vector Machines achieved moderate accuracy (0.68-0.76), while Random Forest reached 0.91 accuracy. Boosting models, including XGBoost, LightGBM, and CatBoost, improved performance to 0.94-0.95 accuracy. Among DL approaches, ANN achieved 0.93 accuracy. The proposed ANN-DFM outperformed all baseline models, achieving 96.65% accuracy and an F1-score of 0.97, while demonstrating stable early-quarter predictions. The findings highlight the effectiveness of multimodal feature fusion for enhancing predictive accuracy and supporting early educational interventions.
This study presents a Collaborative Virtual Reality Learning Environment (CVLE) on coral reef ecosystems, designed for junior high school students to learn through hands-on experience (Learning by doing). The proposed system uses Virtual Reality technology (VR) for the purpose of developing a supplementary learning tool, experienced through the Oculus Quest 2. This VR-based learning system uses Game-Based Learning (GBL) through a mission system, developed with the Unity game engine and Normcore SDK to support collaborative interaction via voice chat. The learning scenario begins with an exploration of organism characteristics, followed by completing three mission levels while observing the ecosystem dynamics. The system features agent-to-agent interactions among organisms, allowing users to observe biological changes and comprehend the roles and relationships of organisms. The results revealed that system usage had a statistically significant impact on participants’ scores (p = 0.0137). Additionally, the VR application rendered at a frame rate of 45 to 70 fps, which is sufficient for system usage. Participants expressed high satisfaction regarding both system usability (4.06/5) and value for specific tasks (4.21/5), with scores ranging from 4 to 5, confirming the system’s effectiveness for the target users. Participants agreed that this proposed system was beneficial in helping to enhance their understanding of the roles and relationships of organisms, and they enjoyed the learning experience.
Traditional vocabulary instruction in primary education often relies on repetition and memorization, which may reduce student motivation and limit immediate feedback during practice. This study presents the design, development, and initial validation of a lightweight educational software application inspired by Hangman to support thematic vocabulary practice through gamification. The novelty of the proposal lies in combining a simple web-based game environment, free technologies, and an iterative user-centered process that integrates Design Thinking and Lean Startup principles for educational software development in resource-constrained contexts. The methodological process includes problem identification, ideation, prototype construction, minimum viable product validation, and continuous improvement. The validation involved 105 participants, including basic-level students, teachers, and parents, selected through purposive sampling. Data were collected using a structured perception questionnaire and open-ended questions. Three educational technology experts provided initial content validation of the instrument and prototype. Quantitative results were analyzed through descriptive statistics and internal consistency estimation, while qualitative responses were examined using ATLAS. ti through coding, category grouping, and triangulation with survey findings. The questionnaire showed high internal consistency, with Cronbach’s alpha of 0.90. The results indicate positive perceived usability, motivational potential, pedagogical usefulness, and user acceptance of the proposed software. However, the findings represent initial validation evidence based on perception and expert judgment, not experimental proof of vocabulary learning improvement. Future studies should apply pre-test/post-test designs, control groups, and longitudinal measures to determine the educational effectiveness of the Hangman-based gamified environment.
A Human-Machine Interface (HMI) is a user interface that allows people to control a machine, carry out a task, and receive feedback from the machine to adjust their control. This paper examines the use of Electromyogram (EMG) signals for machine control via wireless command transmission. The EMG signal includes various muscle responses. The challenge in processing the EMG signal comes from the transient muscle response, which has an unknown arrival time, duration, and shape. The EMG-based Human-Machine interface (HMI) has attracted much attention because it can convert muscle activity into usable control signals. This paper describes the design and implementation of a single-chip platform for an EMG-controlled system to achieve a small, effective and responsive control interface. The development of on-chip real-time processing for EMG-based HMI systems removes the need for an external computer or server. A complete embedded system has been created to control an external car using EMG signals. Different commands are also generated to manage the car's movements. This paper focuses on building a real-time embedded system to control the external car by analyzing hand muscle signals to create commands based on the envelope of both muscle signals. The proposed method for processing EMG signals in a real-time, low-cost system is part of the HMI system, and we discuss its future potential. With the new system, performance accuracy is over 90% in real time.
Flying Ad-Hoc Networks (FANET) has become popular for many applications that require the exchange of reliable, real-time data, such as environmental monitoring or disaster assessment. Most of the missions that rely on FANETs require traffic to meet strict Quality of Service (QoS) requirements (e.g., low latency, high reliability, etc.) and will also have high mobility and often experience sudden changes in their topology. However, almost all routing algorithms currently used in FANETs were not purpose-built to handle these constraints; consequently, they exhibit reduced performance under both conditions. Variation in link quality and intermittent connectivity make maintaining consistent communication more difficult. The proposed QoS-Aware Routing (QAR) framework incorporates application-specific QoS requirements directly into the routing process, enabling route selection to better satisfy the communication demands of diverse FANET applications. The proposed scheme continuously monitors key QoS metrics, such as delay, packet loss, and throughput, and informs routing decisions based on evolving network conditions. QAR will enhance data delivery reliability within the network and increase overall network efficiency by matching real-time QoS measurements to route selection. The simulations show that the proposed algorithm provides better QoS and more predictable communication behaviour than traditional FANET routing protocols, indicating that it can be used in mission-based Unmanned Aerial Vehicle (UAV) applications.
A primary machining operation in today's manufacturing is CNC turning; the optimization of process parameters affects the quality of products manufactured, productivity, and ultimately the cost of manufacturing for all manufactured parts. A systematic review that examines the development and effectiveness of various parametric optimization processes employed in CNC turning was conducted to identify the most effective optimization processes. In contrast to prior reviews, this review will provide an evaluation of optimization processes based on convergence characteristics, computational complexity, scalability, and industrial applications. Both single-objective and multiple-objective optimization methodologies are evaluated in the review, including historical mathematical approaches as well as current artificial intelligence-based methodologies. Metrics such as Surface Finish (Ra), Material Removal Rate (MRR), tool wear, cutting forces, and energy consumption are examined with respect to a variety of workpiece-tool material combinations. Adaptive optimization methodologies in CNC turning that address machining condition variability, real-time adjustment of parameters during machining, and sustainability integration in optimization process frameworks are identified in this review as being major areas of future research. The results of the study indicate a trend from single-objective optimization to multiple-objective optimization in the formulation of optimization processes, and that evolutionary algorithms have demonstrated significantly better performance than other methodologies in addressing nonlinear, multi-constraint industrial problems.
Heterogeneity of cerebral neoplasms, loss of information labeling, and inter-class similarity make it difficult not only to differentiate the tumor clinically but also to establish separation of subtypes, taking Magnetic Resonance Imaging (MRI) into consideration. It presents a neurotumor multiclass classifier, which has a probabilistic soft voting ensemble as the classification method, based on the transfer learning model using pre-trained Convolutional Neural Networks (CNNs), using VGG16, ResNet50 and fine-tuned EfficientNetB0 as the basic classifiers. The experimental data is divided into four different categories: glioma, meningioma, pituitary and no visible tumor. The evaluation is carried out with the use of conventional performance indicators, which include recall, f1-score, confusion matrices, and a micro-averaged ROC analysis. Grad-CAM predicts clinical confidence and interpretability. The fine-tuned EfficientNetB0 model was found to have the highest test accuracy of 73.35%, compared to the equal-weight soft-voting ensemble with test accuracy 70.81%. The results indicate that the ensemble outperformed the ResNet50 and was marginally better than the VGG16, but still not better than the best EfficientNetB0 model. More frequent misclassification patterns, especially class confusion, are compared, and suggestions are made to enhance the sensitivity of the tumor.
Baladweyne City is a city in the central part of Somalia that has experienced disruptions due to flooding over the years. The flooding has always affected the transport and general operations in the Shabelle River floodplain. The main aim of the research is to evaluate the performance of the Support Vector Regression approach in flood forecasting in Baladweyne, with specific emphasis on the dependence of the kernel, the possibility of overfitting and underfitting, and the overall improvement in the accuracy of the results over the benchmark models. The Support Vector Regression approach was implemented using the LIBSVM toolbox with the scaling and normalization of the data and the overall configuration of the input for the models. The results of the Support Vector Regression approach were compared with the results of the transfer function, trend, and naïve persistence models. The results of the research showed that the accuracy of the results of the benchmark models decreased significantly with the increase in the forecast period. For example, the best RMSE of the benchmark models for the first lead time was 170, while the RMSE for the sixth lead time was 1010, showing a deterioration of 494% in the accuracy of the results. On the other hand, the Support Vector Regression approach showed significant improvements in the accuracy of the results over the results of the benchmark models. For example, the Support Vector Regression approach showed improvements of 2.9% in the accuracy of the results for the first lead time and improvements of up to 17.3% for the longer lead times. The results of the Support Vector Regression approach showed that the linear kernels were more robust in the conditions where the rainfall for the future period is not available, while the RBF kernels showed better accuracy when the rainfall for the future period is available. The results of the Support Vector Regression approach showed that the approach is capable of capturing the time of the major flood peaks with minimal relative deviation, while the results of the rainfall-response experiment showed a scaling from low rainfall of 0-4 mm to high rainfall of 50-100 mm.
This paper presents Nested Social Sentiments Classification (NeSS-Class), a sentiment analysis framework that incorporates both primary posts and their nested comments from social media platforms. Unlike traditional approaches, NeSS-Class introduces derived features based on fuzzy string matching to capture subtle textual similarities and to address challenges arising from complex user interaction behaviors in nested discussions. Data was gathered from multiple social media platforms, carefully preprocessed it, and divided into training, validation, and testing sets in an 80-10-10 ratio. Feature stability was evaluated using univariate analysis, and baseline machine learning models were employed for performance assessment. Experimental results demonstrate that Logistic Regression integrated with NeSS-Class significantly improves classification performance, achieving a log-loss value of 0.6553, compared to 0.9099 obtained without the proposed feature set. These results confirm the effectiveness of fuzzy string matching as a feature engineering strategy for enhancing sentiment analysis in noisy, multi-layered social media data.
The COVID-19 pandemic revealed acute flaws in hospitals' decision-making processes, exposing substantial deficiencies in how critically ill patients are managed or in one dimensionalizing complex multidimensional data in a timely manner. Many clinical scoring systems such as the Sequential Organ Failure Assessment (SOFA) and the Acute Physiology and Chronic Health Evaluation II (APACHE-II) are still very widely used based upon their implementation in routine clinical practice, however, both methods are limited by their reliance on a small number of predetermined clinical characteristics as well as not allowing for extraction of high dimensional, data from multiple complex data sources (e.g., volumetric CT). Significant advancements in deep learning techniques have provided significant advancements in terms of both the potential applications as well as how accurately and automatically images can be interpreted. However, most of these methods are limited by only using a single image modality (i.e., either imaging data or clinical data) and therefore provide limited performance since characteristics of either type of data represent only part of the overall information available from the combination of both types of data. To address these limitations, we propose a multimodal deep learning framework for assessing the probability of death in COVID-19 patients that utilizes volumetric CT imaging in conjunction with structured clinical data and integrates a 3D convolutional neural network for CT image feature representation and a dense neural network for clinical representation of clinical data. In order to ensure consistent and reliable data input for our models, we have developed a dedicated preprocessing pipeline for our data that includes lung segmentation and Hounsfield unit normalisation. Results of our experiments completed on Moroccan COVID-19 patient data show that our multimodal approach outperformed the unimodal approach, suggesting that data fusion with respect to predicting risk in a clinical context provides significant benefit. In addition, our proposed framework will be generalisable and may be applicable to many other areas of medicine requiring the integration of multimodal datasets.
