This paper introduces DOA-EEC, energy-efficient clustering for heterogeneous WSN (HWSN), a metaheuristic optimization-based energy efficient clustering algorithm based on dhole optimization algorithm (DOA) for optimal CH selection. Specifically, the method enhances network longevity by utilizing these parameters in a smart way a residual energy, distance between nodes, number of neighboring nodes, multi-hop communication and average remaining energy of the entire network. All these factors place together lead to better CH selection and help also reduce the hot-spot problem, which is a critical issue in case of WSNs. Finally, DOA-EEC significantly outperforms recent optimization techniques with respect to stability period, residual energy, network lifetime, throughput and number of apt CHs through extensive simulations. The proposed algorithm is thoroughly evaluated against a variety of state-of-the-art methods, and statistical tests indicate a considerable performance advantage over the benchmark algorithms used in this work.
To strengthen the health monitoring of the elderly, reduce the risk of falls, and accurately evaluate dynamic standing stability, this study proposes a key feature recognition technology for standing stability based on deep reinforcement learning. First, deploy high-definition devices to capture standing posture images, use BSCB algorithm to repair image defects, then use SIFT algorithm to extract key posture feature points, and construct a time displacement sequence through bidirectional matching strategy. Next, a reinforcement learning recognition framework based on Deep Q-Network (DQN) is constructed, which combines Long Short Term Memory (LSTM) to extract dynamic dependency relationships. Policy networks and reward functions are used to optimize recognition decisions, and experience replay and target network mechanisms are introduced to improve training stability. The experiment shows that this method has the recognition sensitivity can reach 0.95 under optimal conditions and remains above 0.91 in low-light scenarios, strong environmental adaptability and recognition robustness, providing technical support for predicting and intervening in the risk of falls in the elderly.
Currently, images are commonly used, and the number of metaphors, photos, and graphics being created is accelerating. The large volume of data requires a data outsourcing service, such as cloud computing and storage. In addition, it needs a content-based retrieval and exploration solution. The data has been enlarged enormously; however, content-based image retrieval was quite a tiresome job. Image Retrieval is vital because images are used in diverse applications, such as historical research, biodiversity, fingerprint identification, crime prevention, information systems, and medicine. The content-based image retrieval (CBIR) model is usually employed in these cases. From a vast dataset, the CBIR gathers images similar to the query image and removes less beneficial features, then delivers the resulting image as the query image. Next, it compares and matches these features with dataset image features and examines them with parallel features. This paper proposes an Enhanced Content-Based Image Retrieval Using Fusion of Feature Representation with Optimised Image Similarity Measures (CBIRFR-OISM) approach. The main objective is to effectively enhance the system’s capability to retrieve visually and contextually similar images. Initially, the CBIRFR-OISM approach applies image pre-processing using the Gaussian filter (GF) to accentuate edges and structural details. Next, a dual-path deep feature extractor is executed to obtain both local and global representations. The ResNet-50 model captures intricate local patterns, the Vision Transformer (ViT) extracts global contextual representations, and a heuristic search algorithm is employed to select the finest feature maps. The optimal features are then combined through an attention-based fusion model to form a fused feature vector. For image similarity measurement and retrieval, the Manhattan Distance metric is employed to compare feature vectors in the multi-dimensional space. The empirical results of the CBIRFR-OISM method are examined on Corel images, natural images, and the Fruits-360 dataset. The experimental validation of the CBIRFR-OISM method achieved superior precision of 82.32
Depressive disorder is a frequent and debilitating mental illness that must be identified early and monitored on an ongoing basis to be managed. The emergence of speech as a potential non-invasive behavioural biomarker for depression assessment is related to the presence of prosodic, temporal and voice-quality features of depression that can be detected. Current depression detection methods based on speech, however, rely on handcrafted acoustic features and/or black-box deep learning models, have limited usability across different datasets, and are mostly limited to binary screening rather than clinically relevant severity estimation. To overcome these drawbacks, this paper introduces the DepressSenseAI framework and the proposed VoxDepressNet model for effective voice-based depression assessment. It adopts a hybrid fusion strategy that combines self-supervised speech representations with interpretable acoustic and prosodic features. Attention-guided segment-to-session aggregation is used to preserve long-range temporal dependencies and highlight speech segments useful for diagnosis. In addition, a multi-task learning (MTL) approach is used to jointly learn the depression classification and severity estimation tasks when clinicians provide severity annotations, and a domain-adversarial learning (DAL) approach is applied to reduce dataset-specific information in the shared representation and improve robustness across heterogeneous speech corpora. Detailed experiments were conducted on a variety of public speech databases, including ADC, EATD, and PDCH. VoxDepressNet outperformed strong deep learning baselines, achieving up to 0.86 UAR for depression classification and 2.85 MAE for estimating depression severity from clinician-rated HAMD-17 scores. Cross-dataset experiments across diverse language and recording conditions confirmed this. The outcomes indicate the feasibility of using DepressSenseAI for scalable depression screening and mental health assessment of depression severity. The implementation code and experimental configurations are publicly available at https://github.com/mnaveed07154/DepressSenseAI . Not applicable.
Decoder comparisons in typed knowledge graph (KG) link prediction often confound relation specificity, off-diagonal interactions, and directionality. We separate these factors with four seed-matched decoders applied to GCN, GraphSAGE, and GAT encoders: a shared inner product (IP), a relation-specific diagonal bilinear decoder (RD), a symmetric full bilinear decoder (SB), and an unconstrained directed bilinear decoder (D). The study uses a regulation-grounded coal-mine safety KG with 5,093 nodes and 19,314 canonical directed edges. All 120 decoder–encoder–seed cells use the same ten non-adjacent seeds, type-constrained splits, fixed negatives, and training budget. Two-sided paired AUC tests use pre-specified Holm families, with full filtered MRR and Hits@1/3/10 reported descriptively. RD produced no detectable AUC improvement over IP for any of the three encoders under the matched ten-seed design. SB improved on RD for GraphSAGE ( Δ AUC =0.0254 , Holm p=2.15×10^-5 ) and GAT ( Δ=0.1667 , Holm p=3.24×10^-10 ), but not GCN. Releasing the SB symmetry constraint and enlarging the direction-sensitive score space (D - SB) increased AUC for GCN ( Δ=0.0423 , Holm p=4.77×10^-5 ) and GraphSAGE ( Δ=0.0116 , Holm p=0.00251 ), whereas the GAT contrast was not detected. Thus, no RD-over-IP improvement was detected, while the contributions of symmetric off-diagonal interaction and of releasing symmetry while enlarging the direction-sensitive score space depend on the encoder. A deterministic semantic hard-negative stress test, all-family filtered ranking, and a frozen-embedding symmetric/skew diagnostic support but do not enter the confirmatory tests. Candidate discovery uses threshold-free top-five within-family ranking. These results support a narrow conclusion: decoder function class matters in this typed transductive setting, but no decoder is universally superior.
In large-scale directed multi-criteria weighted graph systems, predicting the cost of possible paths or operations is a significant challenge, particularly in areas such as logistics, resource allocation, and workflow management. The non-linear connections and structural information present in such graph data are difficult for current cost prediction methods to handle, which leads to lower accuracy. To improve cost prediction, we suggest a novel method that makes use of graph attention networks (GAT). Our approach uses a multi-head attention mechanism to selectively aggregate important adjacent information, allowing the model to learn feature-rich, localized representations for every node and edge in the weighted graph structure. In particular, the GAT layers are implemented to reflect how the total cost of a directed path or sub-graph operation is influenced by the weights of nearby edges. Our GAT-based model offers a more reliable and accurate framework for cost prediction in dynamic environments. Our experimental results show that the predicted paths match the ground truth paths selected by the Dijkstra algorithm.
Edge-cloud computing infrastructures are becoming a mandate in almost all the possible application areas due to exponential rise in the computational demands over the network. Therefore, adaptive load balancing and workload scheduling is critical in edge-cloud computing for maintaining Quality of Service (QoS) and system agility. To acquire an intelligent task scheduling and load distribution across heterogeneous computing environments, a novel AI-enabled predictive approach is presented in the proposed study for dynamic scheduling framework that integrates a Jacobi Identity-Based Deep Neural Network (JI-DNN) with a Multi-Criteria Grey Wolf Optimizer (MC-GWO). The proposed framework utilizes Jacobi identity properties for enhancing the network’s learning efficiency, stability, and convergence. The JI-DNN is employed for predicting task-specific execution times and heterogeneous resource requirements during run time task allocation in the system and the MC-GWO algorithm adapts the hunting behaviour of grey wolves for optimizing the scheduling decisions based on variant metrics like latency, energy usage, load balance, and overall outcome. The integration of predictive intelligence and multi-objective optimization helps the framework to adapt dynamically to real-time workload variations. Simulation result findings proves the proposed system’s effectiveness, highlighting the crucial improvements in reducing makespan, enhancing energy efficiency, and optimizing resource utilization as compared to the traditional static and single-criteria scheduling strategies. Clinical trial The study is not a clinical trial thus any registration details were not applicable.
Poultry is a major source of food, and growing demand for animal-based products has driven agricultural industries to increase production. However, this expansion has also led to a significant rise in the spread of infectious diseases. Several limitations faced by conventional methods include limited visual indicators, inability to capture complex interactions and limited integration of modern technology. The advancement of modern technology in the poultry industry helps to monitor and track the health of poultry chickens. The early detection of poultry diseases is essential for sustainable poultry farming, reducing poultry losses, and preventing the spread of zoonotic diseases to humans. In this work, a SpinalNet Fusion Recurrent Neural Network (SPFRNN) model is proposed for poultry disease classification based on Deep Learning (DL). The main objective of the proposed research is to design an enhanced deep learning-based poultry disease detection system that enables early and accurate classification of diseases, thereby reducing mortality rates, minimizing economic losses, and preventing the spread of infections. At first, the Internet of Things (IoT) is simulated, and images are collected from IoT nodes. Then, routing is performed at Base Station (BS) utilizing Proposed Jellyfish Search Honey Badger Optimization (JSHBO), whereas the optimal path is predicted by routing based on fitness parameters, such as energy, distance and delay. At BS, poultry disease is detected and classified. Initially, the input image is sent for pre-processing, which is performed by Anisotropic Filtering. Then, the disease area is segmented by Psi-Net, which is followed by Image augmentation. Moreover, suitable features, like Speeded Up Robust Features (SURF), Local binary pattern (CLBP), and Feature Local binary pattern (FLBP), along with statistical features are extracted in the feature extraction stage. Finally, poultry disease is detected by implementing a devised SPFRNN that integrates SpinalNet and Recurrent Neural Network (RNN). The proposed framework performs multi-level classification, where the first level identifies whether the poultry sample is healthy or diseased, and the second level classifies the detected diseased samples into specific disease categories, namely Coccidiosis, Salmonella, and Newcastle disease. Comparative evaluation demonstrates that the proposed SPFRNN model achieves superior performance, exhibiting improvements of 11.476
Forecasting financial markets remains difficult because price series are noisy, nonlinear, nonstationary, and subject to structural breaks. This structured review synthesizes empirical machine learning (ML) and deep learning (DL) research on equities, indices, commodities, foreign exchange, and cryptocurrencies, covering classical ML, recurrent networks, CNN-based models, attention and Transformer architectures, multimodal and graph-based systems, and reinforcement-learning approaches. Because studies differ in task formulation, horizon, data frequency, target variable, feature set, validation design, and metric choice, the review combines qualitative synthesis with structured quantitative evidence mapping rather than a single pooled effect size. The paired-error aggregation identified 17 peer-reviewed studies and 47 same-dataset, same-horizon proposed-versus-baseline error comparisons. On a study-level median basis, this subset showed a median relative error reduction of 20.3
Artificial intelligence (AI) tools such as chatbots have become an inseparable part of education with the advancements in large language models. As more studies have been conducted, researchers and practitioners may have challenges in finding the gaps in the literature and/or deciding on which tools are beneficial or not for instruction. These challenges can be addressed with bibliometric reviews that can provide a comprehensive overview of the current trends in AI use in ELT. This study addresses that need with a novel methodology by examining the ERIC and Scopus databases. Using computational data analysis tools and natural language processing, we examined 856 journal articles and provided several trends, such as keywords, the most cited papers, subject matrices, and geolocations of the articles, as well as several implications for practitioners. A key result is that the research on the use of AI tools is increasing substantially, with a clear focus on teaching methods and writing instruction. Our study offers valuable information for the researchers and practitioners who seek techniques to integrate AI tools into language instruction.
Accurate melanoma detection is essential for early-stage dermatological diagnosis, as it significantly improves treatment outcomes and patient survival rates. This study proposes a robust stacked ensemble framework that combines transfer learning (TL) with particle swarm optimization (PSO) to enhance melanoma classification performance. Initially, four deep transfer learning models: VGG16, ResNet50, DenseNet121, and InceptionV3 are trained on the Melanoma Skin Cancer Dataset consisting of 10,605 dermoscopic images to generate probability scores for melanoma and benign classes. These probability outputs are used to construct feature vectors for the subsequent stacked ensemble stage. The generated feature vectors are then provided as inputs to multiple meta-classifiers, whose hyperparameters are optimized using the PSO algorithm. PSO explores the hyperparameter search space iteratively, improving candidate solutions based on validation accuracy while utilizing both personal best and global best evaluations. Through this process, a total of eleven ensemble transfer learning (ETL) models are developed. Experimental results demonstrate that ETL11, which integrates VGG16, ResNet50, DenseNet121, and InceptionV3 with Gradient Boosting (GB), achieves the highest accuracy of 92.1
Remote sensing scene classification is essential in various Earth observation tasks. However, its accuracy often declines when images are acquired from different sensors or modalities due to domain shift. The differences in imaging mechanisms, sensor properties, and data distributions create significant discrepancies between the source and target domains, making cross-domain generalization challenging. Though domain adaptation methods have presented encouraging outcomes in reducing these drawbacks, the challenges of efficient feature separation, sensor heterogeneity, and distributional mismatch in multimodal datasets remain unresolved. To overcome these disadvantages, the present paper proposes a hierarchical and context-aware feature extraction framework based on the Swin Transformer for cross-domain remote sensing image classification. The proposed approach employs the Swin Transformer to capture both local and global spatial dependencies, enabling better domain-invariant feature representation across heterogeneous sensor modalities. The proposed framework is evaluated on the Multimodal Remote Sensing Scene Classification (MRSSC2.0) dataset across three difficult cross-domain transfer tasks: VIS→SWI, VIS→INF, and VIS→SAR. A thorough comparison among ten domain adaptation methods is conducted to determine the effectiveness of the proposed feature extraction approach. Experimental results show that the proposed framework is particularly effective when domain differences are high, such as in the VIS→SAR transfer. Further evaluation demonstrates the effectiveness of the proposed feature extraction framework when combined with existing domain adaptation methods, with FixBi achieving 96
Accurate leaf instance segmentation is fundamental to quantifying morphological and structural traits in image-based plant phenotyping. However, substantial variations in leaf scale, shape, and orientation, together with dense overlap and occlusion, often lead to ambiguous instance boundaries. In addition, repeated downsampling and feature reconstruction can progressively erode fine structural details, hindering contour preservation and the separation of adjacent leaves. To address these interrelated challenges, we propose DMSHA-Net, a dynamic multi-scale feature fusion and hierarchical attention network for leaf instance segmentation. Its direction-aware Multi-Scale Feature Aggregation (MSFA) encoder captures complementary horizontal and vertical contextual information across multiple receptive-field scales, thereby improving the representation of diverse leaf morphologies. The Dense Feature Aggregation (DFA) decoder selectively integrates deep semantic and shallow structural features through stage-specific attention mechanisms, where self-attention at coarse resolutions models long-range dependencies, whereas lightweight channel recalibration at high resolutions refines local structures and boundaries. The Learnable Feature Fusion (LFF) module subsequently integrates multi-level semantic and boundary features using normalized learned weights. DMSHA-Net achieves Best Dice (BD) scores of 93.17
Under the background of digital transformation of education, it has become a key issue to cultivate the core competence of digital teaching. However, the traditional digital teaching has limitations in situation creation, cognitive interaction and ability evaluation, and it is difficult to effectively support the cultivation of students’ higher-level ability. Multimodal Large Language Model (MLLM) provides a new opportunity to meet these challenges. Based on this, this study constructs a multimodal cognitive enhancement teaching model, which consists of three interrelated modules: multimodal situation generation, adaptive cognitive interaction and core competence evaluation. In this study, 120 college students participated in a 16-week quasi-experiment. The experimental group used the teaching platform based on MLLM, while the control group used the traditional online platform. The data were analyzed by statistical methods such as covariance analysis. The results show that the experimental group has significantly improved in the four core competence dimensions, and the adoption rate of innovative solutions has increased by 32.85
A large part of the cost of producing Digital to Analog Converters (DACs) is related to testing, due to factors such as long time taken for analog fault diagnosis, increased testing time and expensive equipment being required for testing. Therefore, using Machine Learning (ML) based fault classification offers a promising alternative method for testing DACs compared to traditional testing, as it reduces complexity while also improving fault diagnostic accuracy. This study compares the performance of Back Propagation Neural Network (BPNN), Support Vector Machine (SVM), and Probabilistic Neural Network (PNN) to classify catastrophic faults in both 16-bit Charge Scaling and Binary Weighted DACs that have experienced process variation such as variable threshold voltage, oxide thickness and temperature. Additionally, this paper proposes a framework to eliminate the need for additional test hardware, as the simulated fault signatures are used to develop and validate classifiers for both Charge Scaling and Binary Weighted DACs. The experimental results indicate that the BPNN classified faults with an average classification accuracy of 100
Early and accurate discovery of brain tumours is crucial for timely diagnosis and effective treatment planning. In this work, we put forward an explainable multimodal CNN–Transformer framework, kind of merges MRI and CT cues in a complementary way, using attention-guided multimodal fusion so that tumour classification and localisation can happen at the same time. Instead of most existing multimodal CNN–Transformer methods, which mostly linger on feature merging for classification only, this proposal does a bit more. It uses modality-specific CNN encoders, a Transformer based cross-modal attention module, and several fusion recipes—early fusion, late fusion, and attention-driven fusion. On top of that we do quantitative Grad-CAM checking against expert radiologist annotations, plus a computational efficiency examination, all inside a single coherent architecture. Overall this should boost interpretability and make the approach more clinically usable. For training and evaluation, the framework was tested with a carefully curated dataset of 1,986 patient-level MRI–CT pairs collected from BraTS 2021 and the TCIA Brain Tumour CT Archive. The dataset includes glioma, meningioma, pituitary tumour, and healthy controls. We used a patient-wise split: 70
Vehicular Ad Hoc Networks (VANETs) are a major enabling technology for Intelligent Transportation Systems (ITS), facilitating safety-critical, traffic-efficient vehicle-to-vehicle (V2V) communication. Nevertheless, due to the highly dynamic nature of the vehicular environment, heterogeneous mobility patterns that ensure stable clustering and consistent cluster head (CH) selection pose a challenge. Existing embedding-driven and graph-based vehicle clustering methods rely on localized, one-hop neighbourhood aggregation and assume homogeneous neighbour influence. To overcome these challenges, the paper proposes an attention-guided deep embedding framework that jointly models long-range structural interactions and heterogeneous vehicular influence, rather than restricting learning to immediate neighbours. The proposed algorithm trains latent vehicular representations that understand long-range structural interactions that go beyond immediate one-hop neighbourhoods using Deep Nonnegative Matrix Factorisation (DANMF). To capture the heterogeneous vehicle influence (varying mobility, distance over time of vehicles), the dominant connectivity patterns are highlighted by using an ensemble attention mechanism. This incorporates interaction-factor attention to highlight the high-weighted interaction patterns of latent representation, whereas, through vehicle-level proximity attention, the stable and behaviourally related neighbours are selectively highlighted. Clustering is done according to the learned embeddings, and cluster heads are selected based on the strongest affinity of vehicles to cluster centres, to guarantee that CHs represent varying influence of mobility and structural significance. Extensive simulations reveal that the proposed model outperforms baselines, with a maximum throughput of 7.1 Mbps , CH stability of 90 ms , which translates to 36
The online retail platforms of the last decade have gone through an exponential expansion, with Amazon being a prime example. This expansion has resulted in the generation of a vast amount of user–product interaction data, which traditional single-node systems struggle to process effectively. This research presents a scalable, deployment-oriented e-commerce recommendation system that uses Apache Spark, Hadoop, Pig, Hive, and Mahout within a Dockerized big data ecosystem. The pipeline integrates collaborative filtering using Alternating Least Squares (ALS) and content-based filtering using TF–IDF into a hybrid recommendation engine, and is automated end-to-end for ingestion, preprocessing, modeling, and evaluation, with a Flask web interface for monitoring and configuration. The system is evaluated on Amazon review datasets ranging from a 4.6-million-rating modeling subset to a full ingestion corpus exceeding 40 million records. On a real three-node Apache Spark/Hadoop cluster (HDFS replication factor 3), we empirically demonstrate sub-linear growth of training time with data volume, a measurable multi-node speedup, and fault tolerance under a deliberately induced node failure. The hybrid model improves recommendation quality over the standalone models, attaining an RMSE as low as ≈ 0.94 and Precision@10 up to ≈0.74 across the evaluation datasets. This research provides a comprehensive account of the design, implementation, evaluation, and future directions for scalable big data analytics in recommendation systems.
Fog–edge Internet of Things (IoT) systems support low-latency distributed services but remain vulnerable to coordinated attacks that evade detectors designed for independent events. Existing solutions also tend to separate attack detection, provenance verification, and accountability enforcement, leaving no unified path from relational evidence to auditable response. This study aims to develop an integrated framework that detects coordinated malicious behavior, ranks provenance relevance, verifies evidence, and activates rule-based accountability in resource-constrained edge environments. The proposed Blockchain-Fog Computing Collaborative Framework with Deep Attention-based Collusion Detection and Automated Accountability (BF3-ACDA) framework combines a hierarchical blockchain–fog architecture with an attention-based collusion graph neural network (AttnCol-GNN), whose reputation-aware attention coefficient incorporates behavioral correlation and blockchain-derived trust information. A shared attention representation supports both collusion classification and provenance ranking, while Fog-BFT consensus, Merkle verification, and smart contracts provide tamper-evident recording and severity-based enforcement. Across 30 matched independent runs on the collusion-augmented CICIoT2023 benchmark, BF3-ACDA achieved 96.80 ± 0.23
This longitudinal qualitative multiple-case study examined how 24 undergraduate EFL students at a Jordanian public university experienced generative AI during a 12-week academic writing course. The multi-source dataset included successive drafts, archived ChatGPT interactions, decision logs, reflective journals, interviews, and stimulated-recall sessions. Reflexive thematic analysis was combined with episode-level tracing. Impostor phenomenon was used as a sensitizing lens rather than as a measured participant attribute. The findings showed that AI reduced immediate drafting and language difficulties but sometimes shifted uncertainty toward ownership, legitimacy, deservingness, and fear of exposure. Students did not treat AI as a consistently reliable authority; instead, they accepted, adapted, resisted, and, most frequently, verified its suggestions. Concerns about hidden expectations and imagined judgment were evident before a Kafka-focused mini-unit. After the unit, some students used terms such as hidden rules and invisible judges to express these concerns more clearly. The findings therefore suggest that Kafka provided a shared language for reflection rather than creating these concerns or causing changes in authorship. Overall, structured AI use supported more accountable textual decision-making while also generating new affective and authorship-related pressures.
