Array最新文献

{"title":"HitHire: The future of ethical, fair, and sustainable AI recruitment – A governance framework","authors":"Elham Albaroudi ,&nbsp;Taha Mansouri ,&nbsp;Mohammad Hatamleh ,&nbsp;Ali Alameer","doi":"10.1016/j.array.2025.100592","DOIUrl":"10.1016/j.array.2025.100592","url":null,"abstract":"<div><div>Artificial Intelligence (AI) is transforming recruitment but remains susceptible to algorithmic bias and environmental inefficiencies. This paper presents HitHire, a pilot fairness- and sustainability-aware AI hiring platform tailored to the Saudi Arabian context and aligned with Vision 2030 goals. HitHire integrates large language models (LLMs), adversarial debiasing, Shapley Additive Explanations (SHAP), and real-time carbon tracking to ensure transparent and equitable candidate ranking. Evaluated on 350 anonymized CVs across four job roles (web development, finance, human resources, and data science) using a 70/20/10 train/test/validation split, HitHire achieves notable improvements in fairness metrics—Statistical Parity Difference (SPD) for gender = 0.0156 and Disparate Impact (DI) for nationality = 1.2387—while maintaining strong predictive performance (F1 = 0.96 compared to a baseline of 0.80). The system achieves over a 40% reduction in operational CO<span><math><msub><mrow></mrow><mrow><mn>2</mn></mrow></msub></math></span> emissions, with inference energy consumption of 0.003 kWh per query. In a three-month pilot study involving 23 HR professionals within a large Saudi organization, 87% of participants rated system trust at 4 out of 5 or higher. These findings contribute to national digital ethics strategies such as the Saudi Green Initiative, which emphasizes carbon neutrality and sustainable innovation.</div></div>","PeriodicalId":8417,"journal":{"name":"Array","volume":"29 ","pages":"Article 100592"},"PeriodicalIF":4.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145788016","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"End-to-end MLOps framework for EEG-based epilepsy prediction: Automating deep learning pipelines for scalable healthcare solutions","authors":"Salem Trabelsi ,&nbsp;Salah Gontara ,&nbsp;Khaled Ben Khalifa ,&nbsp;Abdellatif Mtibaa","doi":"10.1016/j.array.2025.100649","DOIUrl":"10.1016/j.array.2025.100649","url":null,"abstract":"<div><div>Predicting epileptic seizures from EEG signals is a critical yet challenging task, especially when transitioning machine learning (ML) models from laboratory settings to clinical environments. Most existing approaches lack adaptability to patient variability or remain confined to experimental contexts, relying on manual pipelines with limited automation. They generally do not support deployment and are seldom validated under real-world conditions, limiting scalability and clinical applicability. To address these limitations, we propose an end-to-end MLOps framework specifically designed for EEG-based seizure prediction. Our solution leverages the CRISP-ML(Q) methodology to manage the full ML lifecycle, including automated preprocessing, cross-validation, patient-specific model training, hyperparameter optimization, scalable deployment, and real-time monitoring. The predictive engine integrates a one-dimensional Convolutional Neural Network (CNN) with a Bidirectional Long Short-Term Memory (BiLSTM) network and an attention mechanism to capture spatial–temporal EEG dynamics. Validated on the CHB-MIT EEG dataset, the model achieved an average accuracy of 99.0% and sensitivity of 99.4% on training and test datasets, demonstrating strong generalization and adaptability to inter-subject variability. Following training, the MLOps pipeline autonomously selected optimal configuration parameters, including processing and model architecture, yielding performance improvements exceeding 15% compared to static baselines. In production, the deployed system remained fully operational under intensive concurrent usage, with consistent inference latency below one second in remote conditions. A user-friendly, real-time dashboard enables continuous monitoring, seizure alarm triggering, and live visualization of model behavior and evaluation metrics. These results confirm the robustness, scalability, and clinical readiness of our MLOps-driven framework for real-time EEG-based seizure prediction.</div></div>","PeriodicalId":8417,"journal":{"name":"Array","volume":"29 ","pages":"Article 100649"},"PeriodicalIF":4.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145788316","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Enhancing resilience through bilinear attention pooled region proposal network for metal sheet grid defect classification","authors":"Shyamala Devi M ,&nbsp;Priya S ,&nbsp;Yuvaraj Natarajan ,&nbsp;Sri Preethaa K.R","doi":"10.1016/j.array.2025.100647","DOIUrl":"10.1016/j.array.2025.100647","url":null,"abstract":"<div><div>Defect detection in metal sheet grids is a critical task in the construction industry, where structural reliability and aesthetic quality are crucial. This paper proposes Bilinear Attention Pooled Faster Region-based Convolutional Neural Network (BF-RCNN) to identify and classify surface defects with high accuracy to enhance the building integrity. The novelty of BF-RCNN model relies on generation of Neuro fuzzy weighted otsu (NFO) from 2,400 metal sheet grid images from the public repository. By combining the idea of fuzzy logic and adaptive thresholding, the visibility of defective regions is enhanced by NFO filtering. CHS images are processed by BF-RCNN that uses ResNet-50 for deep multi-scale feature extraction. To strengthen the model's attention to informative features, a Squeeze-and-Excitation (SE) block is integrated into BF-RCNN, enabling channel-wise recalibration and improved sensitivity to subtle defect patterns. Further, Compact Bilinear Pooling (CBP) is employed to model second-order feature interactions efficiently. Also, the representation of complex textures and defect anomalies are enhanced without increasing computational cost. Experimental evaluations show that proposed BF-RCNN model achieves high accuracy, precision and recall of 99.82, 99.12 and 98.94 respectively against existing CNN models. The proposed BF-RCNN model offers significant practical benefits for real-world manufacturing and quality control processes. It also enables high reliable identification of surface defects in metal sheet grids. Thus, the model reduces manual inspection time, minimizes human error, and ensures consistent product quality. This BF-RCNN presents a scalable and intelligent solution for quality inspection in construction industry with both accuracy and reliability in defect assessment processes.</div></div>","PeriodicalId":8417,"journal":{"name":"Array","volume":"29 ","pages":"Article 100647"},"PeriodicalIF":4.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145788318","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"EdgeSched-DQN: An intelligent deep reinforcement learning-based framework for optimized task scheduling in edge-cloud environments","authors":"Nagendar Yamsani ,&nbsp;P. Chenna Reddy","doi":"10.1016/j.array.2025.100645","DOIUrl":"10.1016/j.array.2025.100645","url":null,"abstract":"<div><div>Edge-cloud computing is providing a fundamental for low-latency and resource-intensive Internet of Things (IoT) applications by creating a green ecosystem for task scheduling and resource utilization. That said, these approaches are not able to efficiently handle dynamically varying workloads, do not place resources optimally and also may not scale in an unpredictable environment. Heuristic-based methods are inflexible, conventional machine learning approach is static, while traditional reinforcement learning (RL) suffers from the issues of action space complexity and real-time deployment. To overcome these constraints, we propose a new Deep Q-Network (DQN)-based task scheduling framework, EdgeSched-DQN. An adaptive pruning principle which minimizes action space dimensionality and balances resource allocation with desired levels of system balance, and efficiency, has been integrated into the framework. An optimal task execution, which enforces latency constraints, can be achieved by a tailored reward function that accommodates changing workloads on-the-fly. Abstract It can achieve a 25 % higher reward, 20 % shorter response time, and 18 % higher success rate than state-of-the-art methods. These numbers prove the capability of EdgeSched-DQN as a competent and efficient approach for task scheduling in latency-sensitive applications. It is real-time aware when applied in IoT ecosystems, smart cities and autonomous systems to utilize limited resources more efficiently in edge-cloud environments.</div></div>","PeriodicalId":8417,"journal":{"name":"Array","volume":"29 ","pages":"Article 100645"},"PeriodicalIF":4.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145788319","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"EEG-based prediction of driving performance using spectrograms and vision transformers","authors":"Johan Castro ,&nbsp;Qian Liu ,&nbsp;Camilo E. Valderrama","doi":"10.1016/j.array.2026.100713","DOIUrl":"10.1016/j.array.2026.100713","url":null,"abstract":"<div><div>Cognitive lapses during driving are a leading cause of road accidents, underscoring the need for systems capable of online monitoring of attention and performance. While previous studies have used EEG to assess driver fatigue or focus, most approaches rely on raw signals or handcrafted features, limiting their ability to capture two-dimensional patterns underlying driving performance states. To address this limitation, we propose a deep learning framework that integrates advanced EEG preprocessing with a Vision Transformer (ViT) for predicting driving performance. EEG data from 21 participants performing a simulated driving task were cleaned using ICA to remove artifacts, followed by the extraction of event-related desynchronization/synchronization (ERDS) spectrograms to capture spectral dynamics around driving perturbations. The spectrograms were further enhanced using edge-focused image preprocessing techniques and used as input to a ViT classifier. Driving performance was labeled based on reaction times to lateral perturbations and categorized into good or poor performance. The proposed model achieved 81.8% classification accuracy and 89.8% AUC, significantly outperforming models trained on raw EEG signals and handcrafted features. These findings demonstrate the potential of combining EEG-based spectro-temporal representations with transformer architectures for accurate assessment of cognitive states in driving environments, with potential applicability to real-time systems.</div></div>","PeriodicalId":8417,"journal":{"name":"Array","volume":"29 ","pages":"Article 100713"},"PeriodicalIF":4.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146184706","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A hybrid deep learning framework for early detection of developmental disabilities using speech and behavioral biomarkers","authors":"Abdullah Al-Saleh ,&nbsp;Khalil Hamdi Ateyeh Al-Shqeerat ,&nbsp;Ahmad Hamed AL Abadleh ,&nbsp;Ashit Kumar Dutta ,&nbsp;Ghanshyam G. Tejani ,&nbsp;Seyed Jalaleddin Mousavirad","doi":"10.1016/j.array.2026.100691","DOIUrl":"10.1016/j.array.2026.100691","url":null,"abstract":"<div><div>Early and precise recognition of developmental disabilities (DDs) is essential for the timely intervention of preschool-age children. Although various predictive models have been developed in the past, they cannot provide accurate detections and face challenges like high false prediction, limited generalization, etc. To address these issues, a novel artificial intelligence (AI)-based algorithm was proposed for accurate DD detection. The proposed system integrates digital biomarkers from speech and behavioral domains for predicting DDs. Consequently, an Enhanced Gaussian-Based Noise Filtering (EGNF) is applied to speech data, and Z-score normalization is applied to behavioral data for preprocessing, which makes the raw data reliable for subsequent analysis. Further, feature extraction is performed using Convolutional-Bidirectional Long Short-Term Memory (Conv-BiLSTM) for speech data and Graph Attention Network (GAT) for behavioral data. Subsequently, a Hybrid Feature Selection (HFS) method was developed by combining the Harmony Search Algorithm (HSA) and ReliefF ranking for selecting informative and relevant attributes for model training. Finally, a hybrid classification model named DeepSTNet has been developed by integrating MnasNet and Independently Recurrent Neural Networks (IndRNN), which allows the capture of spatial and temporal long-term dependencies within the data, resulting in high detection performances. The proposed framework was implemented in Python and validated using comprehensive data formed by combining the Dysarthria Detection dataset and the Autism Dataset for Toddlers. The execution outcomes highlighted that the developed framework achieved an accuracy of 99.46%, a precision of 97.74%, and a recall of 98.05%, illustrating that this AI-centric system can be equipped by end users, including healthcare professionals and caregivers, with tools for early screening and detection of DDs.</div></div>","PeriodicalId":8417,"journal":{"name":"Array","volume":"29 ","pages":"Article 100691"},"PeriodicalIF":4.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147396117","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Privacy-aware multi-agent deep reinforcement learning for ethical coordinated control in smart regional grids","authors":"Kexin Zhang,&nbsp;Lingwei Kong,&nbsp;Kun Zhang,&nbsp;Chenxi Huang,&nbsp;Yaqin Kuang,&nbsp;Xiaoming Huang,&nbsp;Rongqiang Feng","doi":"10.1016/j.array.2025.100585","DOIUrl":"10.1016/j.array.2025.100585","url":null,"abstract":"<div><div>With the widespread integration of distributed energy resources (DERs), regional power systems are increasingly reliant on coordinated control among transmission, distribution, and microgrid layers. Traditional control strategies face significant challenges in managing the heterogeneity, volatility, and dynamic load responses inherent in such systems. To address these issues, this paper proposes a privacy-aware and ethically aligned deep reinforcement learning (DRL) framework for optimizing multi-agent coordinated control across hierarchical grid components. The proposed approach constructs a multi-layered state–action space encompassing transmission, distribution, and microgrid subsystems. A multi-agent DRL mechanism is integrated to achieve multi-objective optimization, including load shedding mitigation, overload prevention, reverse power flow control, and voltage stability. Importantly, our design incorporates privacy-preserving training protocols and explainable decision-making modules to ensure transparency, accountability, and secure deployment in critical infrastructure settings. Extensive simulations across diverse operational scenarios demonstrate the strategy’s superior responsiveness, improved renewable energy utilization, and robustness under uncertain conditions. The framework not only enhances operational efficiency but also aligns with emerging global demands for secure, transparent, and ethically governed AI deployment in smart grids. This study provides a novel pathway toward intelligent, privacy-preserving, and responsible regional grid control.</div></div>","PeriodicalId":8417,"journal":{"name":"Array","volume":"29 ","pages":"Article 100585"},"PeriodicalIF":4.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145735331","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Waypoint-guided trajectory planning for mobile robots using GPT-4.1 mini and ensemble learning-based action prediction","authors":"Abderrahim Waga ,&nbsp;Said Benhlima ,&nbsp;Ali Bekri ,&nbsp;Fatima Zahrae Saber ,&nbsp;Jawad Abdouni ,&nbsp;Toufik Mzili ,&nbsp;Ahmed Regragui","doi":"10.1016/j.array.2025.100636","DOIUrl":"10.1016/j.array.2025.100636","url":null,"abstract":"<div><div>Trajectory planning is critical to autonomous navigation systems, working in conjunction with perception, localization, and obstacle avoidance. Traditional path planning algorithms often struggle in large or complex environments due to extensive memory usage and long computation times. In this paper, we propose a hierarchical planning, a multi-level approach where a high-level planner sets general goals for a low-level planner to execute, framework that combines the reasoning capabilities of a large language model (LLM) with the efficiency of a machine learning-based local planner. The LLM acts as a high-level planner by suggesting intermediate waypoints that guide the robot toward its goal. A machine learning-based trajectory planner then uses these waypoints to compute feasible and efficient paths at the local level. This approach significantly reduces the number of states explored during planning and accelerates decision-making. To validate our method, we tested it in 100 simulated environments of varying difficulty levels (easy and hard). The results show that our approach reduces the explored space by 73.2%, 96.9%, 91.6%, and 77.4%, and the length of trajectory required to reach the goal by 5.9%, 5.7%, 2.69%, and 21.1%, respectively, when compared to methods such as A*, Dijkstra, as well as other advanced methods such as an LLM-assisted A* and an improved A* algorithm.</div></div>","PeriodicalId":8417,"journal":{"name":"Array","volume":"29 ","pages":"Article 100636"},"PeriodicalIF":4.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145735329","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Research on different automatic segmentation methods for color cascading framework in detecting malaria infection","authors":"Cucun Very Angkoso ,&nbsp;Yonathan Ferry Hendrawan ,&nbsp;Ari Kusumaningsih ,&nbsp;Achmad Bauravindah ,&nbsp;Rima Tri Wahyuningrum ,&nbsp;Deshinta Arrova Dewi","doi":"10.1016/j.array.2025.100658","DOIUrl":"10.1016/j.array.2025.100658","url":null,"abstract":"<div><div>Malaria remains a global health challenge due to its significant mortality and morbidity worldwide. Our study presents a comparative analysis of five automated segmentation models such as Fuzzy C-Means (FCM), K-Means, Gaussian Mixture Model (GMM), Entropy Filtering, and Watershed, for detecting malaria infection in microscopic blood smear images. The study uses dataset of 574 images, consisting of four Plasmodium species.</div><div>The study uses dataset of 574 images, comprising four Plasmodium species (P. falciparum, P. malariae, <em>P. ovale</em>, and <em>P. vivax</em>) across three developmental stages: trophozoites, schizonts, and gametocytes. Preprocessing techniques used is namely Color Cascading Method, including RGB normalization, gamma correction, noise reduction, exposure compensation, and edge enhancement, were employed to optimize image quality prior to segmentation. Experimental results demonstrate that FCM consistently outperforms other methods, achieving an average accuracy of 98.26 %, specificity of 97.91 %, and sensitivity of 98.61 %. The robustness of FCM in detecting faint and overlapping structures highlights its suitability for automated malaria diagnosis systems, particularly in resource-limited settings. While Entropy Filtering shows promise as an alternative, its inconsistent performance across different lifecycle stages limits its practical utility. The findings of this study provide a foundation for developing accurate and accessible automated diagnostic tools to enhance malaria detection and support global eradication efforts.</div></div>","PeriodicalId":8417,"journal":{"name":"Array","volume":"29 ","pages":"Article 100658"},"PeriodicalIF":4.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145921258","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Green AI techniques for reducing energy consumption in AI systems","authors":"Sunawar Khan ,&nbsp;Naila Sammar Naz ,&nbsp;Tehseen Mazhar ,&nbsp;Muhammad Usman Tariq ,&nbsp;Tariq Shahzad ,&nbsp;Sghaier Guizani ,&nbsp;Habib Hamam","doi":"10.1016/j.array.2025.100652","DOIUrl":"10.1016/j.array.2025.100652","url":null,"abstract":"<div><div>This systematic review synthesizes current evidence on energy-reduction techniques across algorithmic, hardware, and infrastructure layers of AI systems. Model compression and knowledge distillation (e.g., DistilBERT) deliver ∼60 % faster inference with ∼40 % fewer parameters while retaining ∼97 % of baseline performance. Low-precision computation (quantization) yields up to ∼50 % energy reductions, and architecture-level strategies—such as neural architecture search and depthwise-separable convolutions in MobileNetV2—significantly lower compute and memory demand. Specialized accelerators (TPUs) and neuromorphic hardware further improve efficiency, while data-center measures (advanced cooling, virtualization, renewable integration) reduce system-level consumption. For generative-AI workloads, distillation, quantization, efficient architectures, and accelerator-optimized inference remain the primary pathways to lowering both training and inference energy. Across studies, recurring gaps include inconsistent energy-metric reporting, limited standardized benchmarks, and a dominant focus on accuracy over efficiency. Regulatory progress is uneven: the EU has introduced stronger transparency requirements, whereas comparable obligations are not yet global. Review limitations include heterogeneous methodologies and incomplete transparency artifacts, which restrict cross-study comparability. Future research directions include algorithm–hardware co-design, neuromorphic methods, energy-harvesting AI devices, improved data-center operations, and explainable-AI tools to support reliable, energy-aware deployment at scale.</div></div>","PeriodicalId":8417,"journal":{"name":"Array","volume":"29 ","pages":"Article 100652"},"PeriodicalIF":4.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145921200","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}