Fusion: Practice and Applications

Education in the twenty-first century is undergoing a profound transformation fueled by artificial intelligence (AI), the Internet of Things (IoT), and intelligent systems. Traditional and even early digital learning systems often relied on standardized pathways, limiting personalization and reducing learner engagement. To overcome these limitations, this study proposes and evaluates an IoT-enabled adaptive learning framework that integrates educational data analytics with intelligent algorithms to deliver real-time, personalized pathways for learners. Education in the twenty-first century is undergoing a profound transformation fueled by artificial intelligence (AI), the Internet of Things (IoT), and intelligent systems. Traditional and even early digital learning systems often relied on standardized pathways, limiting personalization and reducing learner engagement. To overcome these limitations, this study proposes and evaluates an IoT-enabled fusion-based adaptive learning framework that integrates educational data analytics, ensemble learning, and multi-modal intelligent algorithms to deliver real-time, personalized pathways for learners. The fusion of diverse data sources—ranging from quiz interactions and engagement logs to contextual signals from IoT devices such as smart sensors and wearables—ensures robust, context-aware decision-making. Experimental results using Kaggle datasets demonstrate that Random Forest outperforms XGBoost, with an accuracy rate of 87% and balanced F1-scores. This study shows how AI–IoT fusion can create equitable, eco-friendly, and inclusive learning spaces.

Black fungus disease (mucormycosis) has emerged as a critical health threat, particularly during the COVID-19 pandemic, where immunosuppressed individuals have shown increased susceptibility to opportunistic fungal infections. This study presents a deep learning framework for the automated detection of mucormycosis infections from clinical imaging data. We propose a lightweight yet high-accuracy framework for image-based detection of mucormycosis that couples a pretrained MobileNetV2 backbone with a compact classification head whose key hyperparameters are tuned via Salp Swarm Optimization (SSO). The pipeline standardizes inputs to 224×224 RGB with ImageNet normalization, uses MobileNetV2 as a frozen feature extractor, and lets SSO search the head width uuu, dropout ppp, and learning rate η\etaη under early stopping. On a curated binary dataset (2,991 training / 747 validation images), the SSO search reached a peak validation accuracy of 99.87%, and the final model retrained with the best setting achieved 99.73% validation accuracy. The classification report shows near-perfect performance (diseased: precision/recall/F1 1.00; normal: precision/recall/F1 0.99), with an error rate of ≈0.27% (2/747) reflected in the confusion matrix. Against strong baselines—CNN (90.5%), VGG16 (95.0%), VGG19 (89.3%), InceptionV3 (97.9%)—MobileNetV2 + SSO ranks first while remaining computationally efficient. Grad-CAM visualizations confirm attention on peri-orbital and peri-lesional structures, supporting clinical plausibility. These results indicate that SSO-tuned MobileNetV2 offers state-of-the-art accuracy, interpretability, and deployment readiness for rapid mucormycosis screening.

Remotely sensed images have played a valuable role in several applications such as image classification, feature extraction, land cover monitoring, and others; thus, the need for high-resolution satellite images has become necessary and essential. In order to produce images with very high spectral and spatial resolution, the pan-sharpening techniques—, which are regarded as a subset of data fusion techniques—combine the color information of the multispectral image from the same scene with the distinct geometric features of the panchromatic image. This work conducts a comparative analysis of four pansharpening methods (Gram, HIS, Brovey, and PC) specifically applied to Landsat 7 images, providing a thorough evaluation across multiple performance metrics. Also we introduce and apply performance metrics that not only measure quantitative accuracy (like RMSE and RASE) but also assess the preservation of spatial details, offering a more holistic evaluation of pansharpening techniques. The qualitative and quantitative results indicate that both GS and IHS techniques have accurate performance.

The main goal of this paper is to obtain a special form of asymptotic solutions to the scalar version of the two body problem whenever the two bodies collide on the real line at the collision time. It has been shown that the desired asymptotic solution maintains certain properties when t approaches the collision time. However, it is not easy to Handel such a mission without the employment of successive approximations technique. The successive approximations technique has been modified and adjusted to serve as the main tool in the process of obtaining such solution. Moreover, it has been shown that the series of successive approximations converges absolutely and uniformly to a continuous function that approaches to 0 when t attains the collision time in a certain interval. The problem of one dimensional collision between the two bodies has been solved asymptotically at the collision time.

Heart disease is a severe hazard to the public's health and safety because of the high rates of disability and mortality it causes. Accurate disease prediction and diagnosis are more critical than ever in this era of earlier illness prevention, faster disease detection, and earlier disease treatment. Artificial Intelligence (AI) and the Internet of Medical Things (IoMT) have made it possible to detect, forecast, and diagnose cardiovascular disease more precisely. However, the bulk of these prediction models can only state whether a person is sick; they cannot and do not forecast the severity of the ailment. We present a machine-learning-based technique for predicting cardiovascular disease. Using this strategy, we hope to perform binary and multimodal classifications at the same time. To get things started, we will go through the fuzzy-adaboost approach, which will serve as the foundation for the rest of our work. By combining fuzzy logic and the Adaboost method, this method aims to increase the number of applications that can use binary classification prediction to simplify data analysis. If it is completed, both objectives will be met, and we will eliminate overfitting by merging bagging and fuzzy adaboost into a single approach. It is the ideal solution to the challenge we are currently facing. Because it has a separate classification for the severity of the presentation of heart disease, the bagging fuzzy adaboost can be used for multiclassification prediction. This is because Adaboost's assessment of the severity of the observed heart disease presentations is unclear and imprecise. The results of the experiment reveal that, in addition to a wide range of other classes, the Bagging-Fuzzy-Adaboost can anticipate binary data accurately. When compared to traditional procedures, it is evident that this has significant advantages.

Nowadays, virtual reality (VR) and immersive environments are research fields used in various educational and scientific areas. Immersive digital media desires new techniques for its immersive and interactive features it implies the model of new relationships and narratives with users. VR and technologies related to the virtuality sequence, like digital and immersive environments, are developing media. 3D environments generated with VR compatibility can be skilled from a stereoscopic and egocentric view that outperforms the immersion of the ‘classical’ screen-based view of 3D gamed virtual environments. Recent video games have complete, interactive scenes generated with innovative modeling and animation software and provided with hardware speeded-up graphics and physics. Their communication takes place with body-based sensing and commodity 3D motion controllers, like and in certain ways more progressive, than those discovered in conventional VEs do. Currently, artificial intelligence-based deep learning (DL) methods have been progressively applied to identify and assess user immersion levels in VR environments. In this paper, we present an Advanced Immersion Level Prediction Using Ensemble Classification Model and Metaheuristic Optimization Algorithm (ILPECM-MOA) in 3D Games Virtual Environments. This paper aims to develop a predictive model for assessing advanced immersion levels in 3D game virtual environments using behavioral and contextual data. At the primary stage, the data pre-processing stage uses Z-score normalization to transform input data into a beneficial pattern. Followed by, the presented ILPECM-MOA method designs ensemble models such as the temporal convolutional network (TCN) model, sparse denoising autoencoder (SDAE) method, and stacked long short-term memory (SLSTM) technique for the classification process. At last, the Hybrid ebola and Bald Eagle search optimization (HEBEO) approach fine-tunes the hyperparameter values of ensemble methods and results in the superior performance of classification. The effectiveness of the ILPECM-MOA model has been validated by the detailed studies utilizing the benchmark dataset. The mathematical outcome indicates that the ILPECM-MOA approach has improved performance and scalability in terms of various measures over the recent methods.

The advancement of artificial intelligence (AI) in the field of education system has revolutionized traditional education paradigms. The ability of language models to process human language has revolutionized the field of artificial intelligence. The fusion of deep learning and cognitive science is getting attention in the academic system. The absence of structured policies and lack of AI fusion strategies in academics disrupt traditional teaching classrooms resulting in misuse and resistance in adoption of AI. This marks the importance of preparation of AI policies for effective implementation of AI tools in teaching and learning. This paper highlights the importance of framing the guidelines for organized and practical implementation of AI fusion in academics. This study bridges the gap by developing a standardized framework to transform normal classrooms into dynamic data driven platforms promoting professional development for teachers and empowering students with digital literacy and autonomous learning. The study examines predictive performance using deep learning strategies to extract key features of teaching, learning and cognitive and predicts the impact of AI in sustainable teaching.   The highest importance scores range from 0.89 to 0.94, which indicates the importance of selected key features in models’ predictions. The highest mean score of 4.5 of the model establishes satisfaction of teachers and students with policy objectives. The results of the study indicate that integration of deep learning cognitive strategy along with clear policies framework help in achieving higher adoption and performances rates of AI in sustainable classrooms when compared with traditional teaching strategies with minimal AI-integration.

In this work, the researchers presented an ingenious new way to conceal secret messages within images, a practice called steganography. This technique embedded secret messages within images undetectably. To embed the secret data, it applies a mathematical trick called Discrete Cosine Transform (DCT) that is commonly used to compress image files to hide the secret data in areas of the image that are not too complex or too simple. The algorithm adaptively selected embedding locations based on image texture to the appearance of the image, choosing the most appropriate places to hide the secret and the picture to appear normal. This new method of hiding data is more magical and less detectable than older methods, which modify the smallest details of an image (so-called Least Significant Bit techniques). It examines the patterns of the image such as whether it is smooth or has many details and selects obscure, secure locations to conceal the message. They tried this with 1,000 images, and in each image, they embedded a small message (a paragraph of text). The pictures came out great afterwards with just minor adjustments that most people would not have noticed. 95% of the buried messages could be dragged out flawlessly even after the images had been reduced in size with the JPEG. An artificial intelligence-based high-tech detection tool only detected the hidden data half the time 52%, a significant improvement over the older techniques where it located 85 percent or 65% of the secrets.

Arthritis significantly affects mobility and quality of life due to joint inflammation and dysfunction. Its most common type, rheumatoid arthritis (RA), primarily influences multiple joints and tissues, especially in women aged 30–50. Common symptoms include pain, swelling, and stiffness. The growing prevalence of RA, projected to reach 44 million globally by 2045, underscores the need for advanced diagnostic methods. MRI offers detailed visualization of joint structures, essential for accurate diagnosis. However, current grading systems like OARSI and Kellgren-Lawrence are subjective and prone to variability. This study introduces the KL Grading DeepNetX framework, a deep learning-based model for automated RA grading and classification. The approach integrates image preprocessing and segmentation to extract key features such as joint space narrowing and cartilage thickness. Comparative analysis shows that KL Grading DeepNetX outperforms traditional methods with high precision, sensitivity, specificity, and F1-score. This framework enables earlier, more accurate and efficient detection of arthritis using knee MRI images.

Early and precise detection of corn leaf diseases is important for maintaining crop yield and quality. This work suggests a new end-to-end system Hybrid Adaptive Swarm-enhanced Vision Transformer (HAS-ViT) to overcome the limitations of current techniques such as poor accuracy, high computational expense, and overfitting and inefficient feature extraction. The suggested framework combines a three-stage pipeline such as segmentation, classification and optimization to overcome the issues. First, Adaptive Gradient Masking with Color Entropy (AGM-CE) is a novel segmentation technique that isolates diseased areas through an integration of local color entropy and gradient energy in the LAB color space. This guarantees accurate area selection and removal of the background. Then, a transformer model is constructed named Vision Transformer with Enhanced Visual Attention (ViT-EVA). It integrates depthwise attention layers as well as lesion-aware region concentration, enhancing separation of disease classes and model simplification. Finally, Adaptive Bio-Inspired Gradient Tuning (ABGT) optimizer integrates the Bat Algorithm, AdamW and gradient sign flipping for effective learning and convergence. The mechanism speeds up convergence, prevents local minima and maintains exploration exploitation trade-offs at training. The performance of proposed work is measured on a corn disease dataset and performs at 98.1% accuracy and 0.12 loss than conventional and current transformer-based models.

Technological developments have produced methods that can generate educational videos from input text or sound. Recently, the use of deep learning techniques for image and video generation has been widely explored, particularly in education. However, generating video content from conditional inputs such as text or speech remains a challenging area. In this paper, we introduce a novel method to the educational structure, Generative Adversarial Network (GAN), which develop frame-for-frame frameworks and are able to create full educational videos. The proposed system is structured into three main phases in the first phase; the input (either text or speech) is transcribed using speech recognition. In the second phase, key terms are extracted and relevant images are generated using advanced models such as CLIP and diffusion models to enhance visual quality and semantic alignment. In the final phase, the generated images are synthesized into a video format, integrated with either pre-recorded or synthesized sound, resulting in a fully interactive educational video. The proposed system is compared with other systems such as TGAN, MoCoGAN, and TGANS-C, achieving a Fréchet Inception Distance (FID) score of 28.75%, which indicates improved visual quality and better over existing methods.

Breast cancer is the most prevalent cancer-affecting women worldwide and remains a major cause of mortality. Early detection and accurate prognosis are critical to improving survival outcomes. This study introduces a novel predictive model for breast cancer diagnosis that integrates a lazy learning paradigm with the K-Nearest Neighbors (KNN) algorithm, optimized through a Linear Nearest Neighbor (NN) Search technique and the use of Euclidean distance as the similarity measure. The dataset, comprising 4,024 patient records with 15 clinical and demographic attributes, was obtained from a public repository and underwent rigorous preprocessing, including handling of missing values, normalization, and categorical encoding. The classification model was trained and evaluated using 1:9 cross-validation, with K values ranging from 1 to 9 and a constant batch size of 100 to identify the optimal configuration. Among various configurations tested, the model with K=5 demonstrated the highest performance, achieving an accuracy of 88.02%, precision of 0.87, and recall of 0.88. Additional performance metrics such as F-measure, Matthews Correlation Coefficient (MCC), and Kappa statistic further confirmed the robustness of the selected configuration. The proposed model shows superior predictive capability compared to traditional settings and can serve as a decision-support tool for clinicians. The findings suggest that the combination of lazy learning, effective neighbor search strategy, and robust distance metric can substantially enhance the predictive accuracy of breast cancer diagnosis. This study highlights the potential of machine learning-based tools in clinical oncology, offering a data-driven approach for early intervention and patient outcome improvement.

Plant disease detection using deep learning has achieved high accuracy, but traditional centralized training poses significant privacy risks and incurs high data transmission costs. This study presents a privacy-preserving federated learning (FL) framework for plant disease diagnosis that enables decentralized model training across geographically distributed agricultural sites. Rather than transferring raw farm data to a central server, local models are trained on edge devices and share only model updates. To address data heterogeneity from diverse climates, soils, and plant species, we introduce adaptive aggregation strategies that improve model generalization. Furthermore, we incorporate differential privacy and homomorphic encryption to ensure secure model updates and protect sensitive information from potential breaches. Experimental evaluations on benchmark datasets, including Plant Village and real-world field images, show that the proposed FL-based system achieves comparable accuracy to centralized models while significantly enhancing data privacy and reducing communication overhead. The framework maintains over 93% classification accuracy across 38 plant disease categories, with minimal degradation from added privacy mechanisms. Additionally, we analyze the trade-off between accuracy and communication efficiency, demonstrating the method’s practicality in bandwidth-constrained rural environments. The proposed system offers a scalable, secure, and field-deployable solution for real-time plant disease monitoring, supporting the widespread adoption of AI in precision agriculture without compromising data confidentiality.

The task of automatically generating descriptive and accurate image tags has gained significant attention in recent years due to the exponential growth of image data. Traditional methods for image tagging rely on manual annotation, which is time-consuming and subjective. Automated imagine description fills the gap between visual content and human comprehension, making it vital for activities such as information retrieval, editing, and accessibility. The expanding number of unannotated photographs makes manual tagging impossible. This paper provides a deep learning-based system that combines CNNs for feature extraction, RNNs for caption production, and attention techniques to focus on significant image areas. The model uses a sequence-to-sequence architecture to create coherent captions using pre-trained CNN features and attention-enhanced RNNs. Experiments on datasets such as Flickr8k and Flickr30k show higher performance, as evidenced by BLEU, ROUGE, and CIDEr measures. This approach provides a scalable, cutting-edge solution for image captioning, with potential applications in video analysis, enriched language production, and larger datasets.  

Finding jobs in today’s world is similar to finding a needle in the haystack. The modern job-search platforms present a language barrier for native-speakers and inexperienced candidates, making it difficult for them to compete in the job search race. NaukariCraft, a bilingual (Hindi & English) job search platform makes it easy for users to look for jobs, gain industrial insights, save time by finding relevant jobs tailored to skills and resume, building ATS friendly resume, and ATS score analyzer. NaukariCraft provides full guidance to novel applicants helping them find direction towards jobs tailored to their resume. Using advanced technology like Large Language Model (LLMs) and agents, NaukariCraft enhances user experience, improves employability through resume analysis, and reduces application fatigue. This paper outlines the methodology, proposed work, result, conclusion and future development avenues for NaukariCraft.