Information asymmetry and the misreporting of operational data severely challenge contemporary supply chains (SCs), leading to inefficiencies, disruptions and performance deterioration. Traditional anomaly detection techniques typically focus on detecting downstream consequences rather than proactively identifying misreporting behaviors. To address this gap, this study introduces an interpretable AI-driven framework, employing the Isolation Forest algorithm, for the early detection of misreported data in multi-tier SCs. Central to this approach is the introduction of Engineered Cross-Stage Features (ECSFs), designed to capture relational discrepancies across different SC echelons. The framework was evaluated using a synthetic three-tier SC dataset, comprising both standard operational conditions and artificially perturbed data points representing misreporting behaviors and different scenarios of information asymmetry among SC actors. In a comparative assessment against Hotelling’s T2 and an Autoencoder, our ECSF-based Isolation Forest framework demonstrated superior efficacy, highlighting its potential as a practical solution for proactively identifying and mitigating misreporting in complex SC environments.
Hand gesture recognition remains challenging, primarily due to its reliance on large-scale annotated datasets and the limited adaptability of existing models when encountering novel gesture classes. In this work, we propose to apply Adaptive Vision-Language Model (Adaptive-VLM). This lightweight, training-free framework utilizes only one image per class to recognize gestures on the HaGRID benchmark. Built upon the CLIP backbone, our approach incorporates symbolic knowledge-infused prompts, multi-prompt contextualization, and semantic exemplar ranking to improve few-shot generalization. Adaptive-VLM achieves a macro F1-score of 65.75% on the HaGRID test set (540 images) without any parameter fine-tuning, using 18 example images. It significantly outperforms the Random-VLM baseline (59.95%) and a ResNet-18 model fine-tuned for 10 epochs (4.09%) under the same data constraints. These findings highlight the effectiveness of combining structured domain knowledge and guided exemplar selection to overcome data scarcity in low-resource gesture recognition. Adaptive-VLM offers a promising direction for building adaptive and efficient HGR systems, especially in real-world human-computer interaction scenarios requiring rapid deployment with minimal data.
Transatlantic cruises in the northern Atlantic often face medical situations requiring specialized care unavailable onboard, necessitating patient transfers to onshore facilities. The Portuguese archipelago of the Azores, strategically located along their routes, frequently receives such vessels. Currently, ship-to-shore communication relies on email, which is prone to errors and raises data security and privacy concerns. To address these challenges, this article describes the development of a 3-tier web application prototype with a React frontend, Express backend, and PostgreSQL database. This application adheres to the FHIR and openEHR healthcare standards, improving data modeling and creating comprehensive electronic health records. It transcends language barriers and streamlines clinical workflows, promoting seamless data exchange and greater efficiency in patient care. The benefits are substantial: error reduction through comprehensive patient overviews, mitigation of language barrier risks, optimized resource coordination, lowered costs, improved response times, and enhanced experiences for professionals and patients through more accurate, efficient, and secure medical care.
This work proposes an approach for dynamic facial expression recognition to recognize emotions in controlled environments, due to its computational efficiency. The CREMA-D and RAVDESS datasets are used, from which sequences of 100 frames per video are extracted. Preprocessing for geometric features is performed using Face Mesh and facial alignment, while for deep features, face detection, facial alignment, resizing, and center cropping are applied. Geometric features are computed from the internal angles between facial landmarks, and deep features are extracted using MobileNetV2, ShuffleNetV2, and EfficientNet-B0, followed by dimensionality reduction via NCA. Both representations are concatenated and used as input to an LSTM (for CREMA-D) and a BiLSTM (for RAVDESS). The proposed method achieves UAR/WAR scores of 63.68%/63.71% on CREMA-D and 79.50%/80.21% on RAVDESS, demonstrating that the proposed approach is efficient and competitive without relying on architectures with higher computational cost.
Workplace stress, a widespread issue in modern professional environments, significantly increases the potential for errors and accidents. Timely and precise stress identification is vital for fostering a secure and efficient work environment. This research introduces an innovative, comparative-analysis framework designed for real-time stress detection, utilizing Heart Rate Variability (HRV) as a reliable physiological indicator. Unlike standard heart rate measurements, HRV offers a granular view of the autonomic nervous system (ANS) function, enabling accurate stress evaluation. We implement a comprehensive methodology incorporating a refined preprocessing stage—including the removal of outliers, feature selection, and data normalization—along with a comparative assessment of eight deep recurrent neural network (RNN) architectures. These include vanilla RNN, bidirectional RNN (BiRNN), Gated Recurrent Unit (GRU), bidirectional GRU (BiGRU), standard Long Short-Term Memory network (LSTM), bidirectional LSTM (BiLSTM), Peephole LSTM, and Attention-based LSTM, applied to binary stress classification. Utilizing a dataset of 410,322 HRV records from the SWELL Knowledge Work (SWELL-KW) Dataset, our framework demonstrates exceptional performance, with the BiGRU architecture achieving a test accuracy of 99.51%. This study highlights the effectiveness of advanced temporal modeling and comparative analysis in creating robust stress detection systems for various occupational contexts, thereby enhancing workplace safety.
The sparsest packing problem emerges in manufacturing multi-hole extrusion dies to obtain small and precise products in the automotive, aviation, food, and medical industries. The goal is to maximize the minimum Euclidean distance between the objects and between the objects to the boundaries of the container. Additionally, this task might be subject to balancing constraints that determine that the deviation of gravity center of the system should stay within a threshold. We present a novel custom environment that encompasses the constraints present in this task. We experiment with the proposed environment using Proximal Policy Optimization to assess the applicability of reinforcement learning for the sparsest packing problem with circular objects in a circular container. Our results indicate that the proposed agent learns efficiently, demonstrating promising results in both finding feasible solutions and optimizing the placement of objects. Our approach not only shows the potential of reinforcement learning for solving the sparsest packing problem but also provides insights into its effectiveness in environments with complex spatial and balancing constraints.
This paper proposes a hybrid access control system that integrates the usability of Web2 authentication (Google Login) with the transparency and integrity of Web3 technologies (blockchain and smart contracts). The system enables users to authenticate via their existing Google accounts without managing crypto wallets or private keys, while access permissions are securely recorded on-chain through smart contracts. To ensure cryptographic key security without relying on a centralized authority, the design incorporates Distributed Key Management (DKM). This approach addresses the challenge of balancing usability with verifiability in data access control. By embedding decentralized guarantees within a centralized web service interface, the system enables practical and transparent access control. The proposed architecture demonstrates the potential for a general-purpose, auditable module that facilitates user-consented data sharing with third parties.
This paper introduces a multi-method explainable artificial intelligence framework designed as a foundational step toward European Union AI Act-compliant automated ICD-10 coding. The framework integrates three complementary explainability methods: label-wise attention, SHAP and case-based reasoning. Unlike existing approaches that typically employ single explanation methods, this new framework creates a comprehensive implementation of multiple explainable techniques while implementing stratified evaluation across different data complexity levels. The architecture leverages a transformer-based model with label-wise attention aggregation. While current performance levels require further development before industry deployment, preliminary results demonstrate competitive performance with micro-F1 score of 0.565 and explanation coverage of 87.1%, establishing critical infrastructure for regulatory-compliant explainable AI in healthcare.
Against the backdrop of accelerating globalization and increasingly fierce corporate competition, financial management, as the core of corporate governance, is facing growing complexity. Traditional approaches fall short in handling massive financial data, volatile markets, and elevated decision demands, while AI and big data technologies offer robust support for intelligent decision-making. This paper presents a hybrid framework integrating AIGC with financial expertise, leveraging NLP and knowledge base retrieval to enable automated financial knowledge analysis. Experimental results validate its high accuracy in knowledge matching, addressing traditional limitations and advancing the intelligent transformation of financial management.
This paper proposes an integrated mixed-integer programming model, termed the Single-Track Railway Cyclic Scheduling Model (SRCSM), for constructing optimized periodic timetables (operating on a recurring 24-hour cycle) for bidirectional single-track railway systems. The SRCSM enhances operational efficiency by precisely considering train arrival/departure times, safety headways, platform track allocations, and meet/pass operations. Validation using real-world data demonstrates its practical applicability, flexibility, and scalability for dynamic timetable optimization, offering a robust tool for improving operational efficiency and safety in single-track railway operations.
This study addresses the scheduling problem in a single-machine system with variable processing speed. The performance of the Earliest Due Date (EDD) rule, operating at a fixed maximum speed, is compared with a Reinforcement Learning (RL)-based approach capable of dynamically adapting both processing speed and job selection. The analysis considers both temporal performance (tardiness) and sustainability (energy consumption). Simulation results, supported by statistical analysis, show that the RL approach achieves a significant reduction in total energy consumption compared to the EDD rule. Regarding temporal performance, analysis of variance reveals that the chosen policy significantly affects the number of tardy jobs, with an effect that depends on the tightness of job deadlines, whereas the overall average tardiness across all jobs does not differ significantly between the two policies. These findings highlight the potential of adaptive RL-based policies for more sustainable resource management, demonstrating the ability to achieve substantial energy savings while maintaining competitive temporal performance and dynamically managing the trade-off according to deadline pressure.
Planning maintenance in manufacturing systems is challenging, especially when dealing with unpredictable machine breakdowns. This paper presents a Deep Reinforcement Learning (DRL) framework to automate and optimise maintenance decisions. Our approach uniquely models a realistic industrial environment where machine failures are stochastic and can occur in succession, a critical factor often simplified in traditional methods. The DRL agents learn to make decisions using local machine data combined with key system-wide performance metrics, enabling a modular yet globally-aware strategy. We benchmarked our DRL policy against a state-of-the-art Metaheuristic Genetic Algorithm (MGA). The results demonstrate that our DRL approach achieves two key advantages. First, it matches the production throughput of the benchmark, particularly under moderate operational stress. Second, it significantly reduces the overall maintenance workload and enhances system robustness against the variability of machine failures. This work highlights the potential of DRL to create intelligent, decentralized maintenance strategies that improve both efficiency and resilience in complex, high-variability industrial environments.
Magnetic Resonance Imaging (MRI) education often suffers from limited access to physical scanners and the complexity of MRI parameter interdependencies. This paper presents a web-based MRI simulator integrated with a knowledge-based AI assistant to enhance medical and radiography training to bridge the gap between theoretical learning and practical experience in MRI procedures. The simulator offers medical and radiography students an intelligent, interactive platform accessible via any web browser. The simulator enables interactive MRI parameter adjustments and real-time imaging feedback, while the AI assistant employs structured knowledge representation and rule-based reasoning to provide personalized recommendations on parameter optimization, artifact reduction, and protocol selection. The system was developed using modern web technologies including Node.js as a backend solution, JavaScript-driven frontend to scalable with smooth navigation and MongoDB for database. The system’s intelligent assistant leverages domain-specific ontologies and rule-based reasoning to deliver personalized recommendations on parameter optimization, artifact reduction, and protocol selection. The usability testing with 30 medical, system development students and instructors demonstrated high satisfaction, achieving average usability and learning effectiveness scores of 85.1% and 84.7%, respectively. These results indicate the system’s potential to bridge theoretical learning and practical skills in MRI education. Future work will expand anatomical coverage and integrate deep learning to further enhance the AI assistant’s adaptability.
With the global emphasis on environmental sustainability, particularly in the context of industrial manufacturing, and the need to comply with government regulations, organizations have been compelled to adopt sustainable manufacturing processes and report their current sustainability levels and strategic goals. The organizations are using multicriteria decision-making frameworks to assess their current sustainability performance and prioritize key elements for strategic decision-making. However, the current sustainability assessment frameworks involve complex and laborious procedures, starting with the collection of data from organizational databases and then computing their sustainability level. In this regard, the study presents a structured application for calculating the organizational sustainability index (SI). The developed application utilizes the Analytical Hierarchy Process (AHP) and an aggregation method to calculate the overall sustainability index. The integration of the application with the organizational database enables seamless calculation throughout the entire process. The application provides a better understanding of sustainability performance and enhances organizational strategic decision-making by prioritizing the most influential sustainability indicators.
Large Language Models (LLMs) have shown increasing potential in automating model-driven software engineering tasks, particularly in generating models conforming to Domain Specific Languages (DSLs) from natural language. While most existing approaches rely on large proprietary models, their high cost and limited deployability hinder broader adoption. In this paper, we evaluate whether open-source LLMs of varying sizes (0.5B to 32B parameters) can generate DSL-conformant models using only few-shot prompting, without any fine-tuning. Our evaluation focuses on key model-driven engineering (MDE) requirements, including syntactic validity, semantic completeness, and inter-model reference consistency. We extend our prior work by moving from generating user interface models (referred to as "UI models" in this paper) over fixed, predefined data schemas ("data models") to generating both the UI and data models entirely from scratch. This shift serves two purposes: first, it highlights the LLM's ability to infer domain-specific relationships and maintain consistency across multiple interconnected models; second, it allows us to generalize earlier findings by testing DSL generation across models of different natures and structural roles. Our structured evaluation combines automatic parsing and expert feedback across 39 LLMs, revealing that several compact models (e.g., , ) approach or match the quality of much larger models. These findings demonstrate the feasibility of using smaller, open-source LLMs for grammar-conformant DSL generation in MDE workflows, offering a cost-effective and deployable alternative to closed LLMs.
This paper presents an intelligent digital concordance system powered by Natural Language Processing (NLP) to advance the study of Arabic scripts and support the development of faith-based intelligent applications. It addresses the limitations of traditional concordance methods by applying advanced computational techniques to analyze the linguistic and thematic structures of Arabic texts. A comprehensive NLP framework was developed, incorporating Arabic morphological analysis, semantic similarity detection, thematic clustering, and cross-referencing algorithms. The system processes the full Arabic corpus (comprising 6,236 verses across 114 chapters of the Quran) using transformer-based models fine-tuned for Classical Arabic, alongside traditional linguistic tools. The proposed framework enables automated indexing, semantic search, and bilingual alignment between Arabic and English texts. Experimental evaluations show strong results, with over 94.7% classification accuracy, 89.3% clustering precision, and a 92% effectiveness rating by domain experts. This research highlights the effective integration of modern NLP techniques for sacred text analysis. By combining linguistic integrity with computational intelligence, the framework offers a robust foundation for faith-aware AI systems and provides scalable, context-sensitive, and semantically rich access to religious knowledge which enhancing academic research and digital scholarship in Arabic script studies.
The emergence of forged engineering degrees submitted to the Board of Engineers Malaysia, fake diplomas issued by a former college CEO, and online syndicates selling counterfeit academic credentials for RM1,500 to RM4,000 clearly demonstrates the failure of current institutional controls and emphasizes the urgent need for a secure and unified verification system. This paper examines the potential of blockchain technology as a transformative solution for academic certificate verification. This aligns with Malaysia’s National EdTech Policy and the 2018 rollout of the blockchain-based eScroll by the Ministry of Higher Education and six public universities, as well as broader national strategies including MyDIGITAL under the 4IR Blueprint and Malaysia Blockchain Infrastructure. The study reviews current implementations and research on blockchain-based systems, including case studies such as CredChain, EduTrust, and UTM-BADVES, which demonstrate practical viability and enhanced security. A detailed examination of the eScroll system, developed by the International Islamic University Malaysia (IIUM), provides insights into a localized application of blockchain for academic credentialing. The eScroll system utilizes a permissioned blockchain, enabling only authorized institution, such as accredited universities and regulatory bodies to access, verify, and update academic credentials securely and transparently. To ensure data integrity and tamper-resistance, the system employs cryptographic hash functions such as SHA-256, which generate unique, immutable digital fingerprints of each credential. While the technology presents clear advantages, including fraud prevention, real-time validation, and cost reduction, challenges remain in terms of scalability, regulatory compliance, and interoperability. The paper concludes by emphasizing the need for strategic institutional adoption, standardized frameworks, and cross-sector collaboration to fully realize the benefits of blockchain in securing Malaysia’s educational ecosystem. Future work include, integration with the Malaysian Qualifications Agency (MQA) databases would allow real-time synchronization and validation of program accreditations and graduate information, enhancing regulatory oversight and ensuring consistency with nationally recognized qualification frameworks. To further enhance adoption and practical utility, blockchain-based credentialing systems such as eScroll can benefit from strategic partnerships with professional networking platforms like LinkedIn and major job portals.
Anomaly detection (AD) is a technique to detect abnormal samples. However, there is a risk that attackers create anomalous AD models. The risk will be higher in federated anomaly detection, where the privacy of training data is protected. To address such a risk, this paper conceptualizes a novel problem, namely anomalous anomaly detector detection (AADD). The idea is to apply AD to AD models. For this purpose, we represent AD models as rankings of normal scores. The hypothesis is that anomalous AD models create different rankings than normal ones. Accordingly, one can transform AADD into an anomalous ranking detection problem. This study combines the nearest neighbor and ranking correlations to detect anomalous rankings. The experiment is conducted with five binary classifications and four one-class classification algorithms. The proposed method can classify AD algorithms with 100% accuracy when all AD models learn the same class. Future work includes feature extraction from AD models and detecting anomalous AADD models.
This paper presents the design, implementation, and validation of the X-Band Radar Information System (XRIS), an on-premises, web-based platform enabling real-time ingestion, processing, visualization, and secure dissemination of rainfall measurements from an X-Band Multiparameter Radar (XMPR). Developed at Universiti Teknologi Malaysia (UTM) Pagoh, XRIS modernizes previously fragmented and manual radar data workflows by automating data ingestion and providing centralized, role-based access for researchers, disaster-response teams, and government agencies. Built using Agile methodology and a modular architecture based on Django, Celery, PostgreSQL, Redis, and RabbitMQ, XRIS reliably processes radar data every two minutes, offering responsive real-time visualizations and secure public access through Cloudflare Tunnel. Comprehensive testing and stakeholder validation demonstrate XRIS as a scalable, sustainable platform that bridges the gap between academic research and operational meteorology. By automating and centralizing legacy workflows, XRIS enhances the accessibility and usability of radar data, supporting national disaster preparedness and advancing hydrometeorological research.
The rapid expansion of healthcare data necessitates scalable, interoperable, and privacy-preserving mobile data management solutions. MedEForm addresses this by integrating the openEHR standard with Google Firebase to deliver a real-time, cloud-based electronic health record (EHR) platform optimized for Android devices. Utilizing Archetype Definition Language (ADL) models from the openEHR Clinical Knowledge Manager (CKM), MedEForm features a GUI generator that dynamically renders context-aware data-entry forms client-side, eliminating the need for app updates. Authentication is securely managed via Firebase Authentication, while encrypted user credentials and EHR data are stored in Firestore, Firebase’s NoSQL document store. A dedicated query module enables role-based execution of single-patient, multi-patient, and cohort-level queries across a 90,000-instance dataset. Firestore’s real-time synchronization, offline support, and fine-grained security rules enforce low-latency access with robust access control. For epidemiological analysis, MedEForm implements anonymization of demographic, clinical, and geospatial data at weekly intervals, storing these de-identified records in a separate Firestore collection. This architecture supports high-quality analytics while upholding stringent privacy guarantees. By unifying open standards, dynamic interface generation, and secure cloud infrastructure, MedEForm offers a modular and interoperable framework for mobile health data collection and analysis.
