Osama AlQahtani, College of Engineering and Computer Science, University of Jazan, Jazan, Saudi Arabia
Ultra-Reliable Low Latency Communications (URLLC) applications in 5G and beyond net- works demand unprecedented performance levels with sub-millisecond latencies and 99.999% reliability. While AI-based traffic classification has emerged as a critical enabler for intelligent network management, existing approaches focus primarily on algorithmic improvements without adequately addressing the prac- tical constraints of edge deployment environments. This paper presents a novel theoretical framework for edge-native URLLC traffic classification that systematically bridges the gap between AI model capabil- ities and real-world deployment limitations. The framework comprises four interconnected components: resource constraint modeling, latency decomposition analysis, reliability-performance trade-off optimiza- tion, and edge-cloud orchestration principles. A systematic four-phase methodology guides practitioners through system characterization, model selection and optimization, deployment strategy determination, and performance validation. Theoretical case study analysis across three representative scenarios—5G base station edge computing, industrial IoT gateways, and vehicular edge computing nodes—demonstrates the framework’s effectiveness in diverse deployment environments. Framework projections indicate potential achievement of URLLC targets with 0.7ms average latency and 99.997% reliability for high-resource sce- narios, and 400μs latency with 99.9993% reliability for constrained industrial applications. The theoretical analysis shows consistent resource optimization potential of 45-78% while maintaining acceptable classifi- cation accuracy. This work provides the foundation for systematic deployment of AI traffic classification systems in edge environments, offering both theoretical rigor and practical guidance for next-generation URLLC applications.
Nikitha Merilena Jonnada, University of the Cumberlands, USA
In this paper, the author discusses about the significance of network security as it continues to grow and how digital infrastructure is becoming an integral part of our daily life. Modern networks are facing more threats like the malware, ransomware, phishing, advanced persistent threats (APTs), and vulnerabilities in the cloud and Internet of Things (IoT) environments. These threats compromise data confidentiality, integrity, and availability, posing risks to individuals, businesses, and the government. This paper presents a comprehensive review of contemporary network security principles, technologies, and practices, while proposing an Integrated Network Defense Framework (INDF) that combines technical, administrative, and policy-driven measures for holistic protection. The paper examines the foundational CIA triad (confidentiality, integrity, availability), common attack vectors, and modern defense mechanisms, such as firewalls, intrusion detection and prevention systems, endpoint detection and response, and encryption methods.
Network Security, Cybersecurity, Zero Trust, Artificial Intelligence, Threat Mitigation.
Amirmasoud Soltanzadeh and Zbigniew Dziong, Department of Electrical Engineering, Montréal, Québec, Canada
The increasing use of sensor-based wireless communication systems, such as WSN, provides numerous benefits; however, some issues exist in their utilization, most notably regarding energy efficiency. In previous times, multiple approaches existed to address the energy issue in WSNs; however, it is necessary to adapt efficient approaches to overcome the remaining issues. This study utilizes an advanced artificial intelligence algorithm, clustering, and a routing process to enhance the performance of WSN intelligently. The algorithm used for clustering is PSO-mutation, which is employed to choose CHs. Golden eagle optimization is used for route optimization within the CHs to minimize energy expenditure and enhance the WSN’s lifetime. Matlab tool to simulate and evaluate with AI techniques based on a genetic and predictive coding theory algorithm, as well as the traditional Leach-CE-based routing protocol for WSNs. Performance metrics include energy consumption, the number of dead nodes, throughput, and delay. The proposed model demonstrates significant improvements over the Leach-CR model, thereby justifying its validity.
Golden eagle optimization (GEO), Particle swarm optimization-mutation (PSO-Mutation), Energy efficiency routing protocol (EERP). Wireless sensor networks (WSNs).
Bornoma Halima, Farouk Hafsa Mu’azu, Okonta Ehijesumuan, Diadie Sow, Ignace Djitog and Ekpe Okorafor, Nigerian British University, Nigeria
Bone marrow diseases are illnesses that affect the bone marrow of an individual. The bone marrow is a mesh-like organ situated inside the bones of a human being entirely in charge of producing blood and all blood components i.e. lymphocytes, erythrocytes, platelets and plasma. Any disease affecting the bone marrow affects the production of blood which can lead to loss of blood or cancers which eventually lead to death. This paper proposes a new web-based application integrated with convolutional neural networks algorithm, a machine learning approach, to automate an early diagnosis of bone marrow diseases without much hassle contrary to common manual processing which is exceedingly labor-intensive and costly. While the dataset was thoroughly examined, features that fit in with patient characteristics living with sickle cell disease in Nigeria were extracted to carry out the analysis. The dataset contains 11 classes of different bone marrow disease cell types, and a number of performance measures such as area under the curve (AUC), precision, and recall were generated and analyzed with the following results 98.38%, 87.12%, and 77.12% respectively. In terms of diagnosing sickle cell diseases with patients in Nigeria, the proposed model surpassed all existing learning models. The resulting model was saved in a specific file format then successfully imported into the developed web portal for instant analysis and wider access by authorized personnel.
Machine Learning, Convolutional Neural Networks (CNN), Bone-Marrow Disease, Hematology, Smear, Prediction, Web Application.
Yousef Mehrdad Bibalan1, Behrouz Far1, Mohammad Moshirpour2, and Bahareh Ghiyasian3, 1 University of Calgary, Canada, 2 University of California, Irvine, USA
Work-in-Progress (WiP) prediction is critical for predictive process monitoring, enabling accurate anticipation of workload fluctuations and optimized operational planning. This paper proposes a retrieval-augmented, multi-agent framework that combines retrieval-augmented generation (RAG) and collaborative multi-agent reasoning for WiP prediction. The narrative generation component transforms structured event logs into semantically rich natural language stories, which are embedded into a semantic vector-based process memory to facilitate dynamic retrieval of historical context during inference. The framework includes predictor agents that independently leverage retrieved historical contexts and a decision-making assistant agent that extracts high-level descriptive signals from recent events. A fusion agent then synthesizes predictions using ReAct-style reasoning over agent outputs and retrieved narratives. We evaluate our framework on two real-world benchmark datasets. Results show that the proposed retrieval-augmented multi-agent approach achieves competitive prediction accuracy, obtaining a Mean Absolute Percentage Error (MAPE) of 1.50% on one dataset, and surpassing Temporal Convolutional Networks (TCN), Long Short-Term Memory (LSTM), and persistence baselines. The results highlight improved robustness, demonstrating the effectiveness of integrating retrieval mechanisms and multi-agent reasoning in WiP prediction.
Predictive Process Monitoring, Work-in-Progress, Retrieval-Augmented Generation, Large Language Models, Multi-Agent Framework
Ye Li, University of Canberra, Australia
Phishing persists because current email architectures cannot guarantee individual-level sender authenticity. Domain-based defenses (SPF, DKIM, DMARC) validate organizational infrastructure yet admit execution paths where messages from compromised accounts are accepted without verifying the named individual. We introduce authenticity completeness, a global property requiring that every execution of the email transmission process terminate either in verified authenticity or explicit rejection, thereby forbidding any form of unverified acceptance. We formalize this property via a state-machine model, prove that domain-level mechanisms are structurally incapable of satisfying it, and present a document-oriented digital signature scheme that uses ephemeral per-message keys. The design satisfies four reinforcing properties—proactive defense, individual-level authenticity, transparency of operation, and localized deploy ability—and we prove that it achieves authenticity completeness. This yields the first rigorous end-to-end guarantee against individual level impersonation within email systems.
Email Authenticity, Phishing, Digital Signatures, Authenticity Completeness.
Anna Forster1, Carlo Lucheroni1 and Stefan Gürtler2, 1University of Camerino, Camerino, MC, Italy, 1University of Applied Science of North-west Switzerland, Olten, Switzerland
The Digital Out-of-Home (DOOH) advertising industry still struggles to achieve precision in getting audience attention focused, relying mainly on location-based targeting that neglects critical consumer context. This paper presents an Ontology-Enhanced Multimodal Targeting System designed to bridge this gap. The core innovation of the proposal is a domain-specific ontology that integrates anonymized mobile Global Positioning System (GPS) data and real-time contextual inputs (e.g., time, weather). This semantic framework enables the operational advertisement targeting system to move beyond data correlation, inferring audience insights that are synthesized into a composite business-centric Recommendation Score (MR). A field experiment validates the system’s performance against a traditional baseline, demonstrating a 220% relative uplift in relevant advertisement impressions. The presented findings establish a quantifiable and privacy-conscious methodology for optimizing DOOH advertising delivery, possibly positioning the proposed ontology-enhanced approach as a foundation for privacy-conscious contextual personalization.
Advertising, digital out-of-home; mobile GPS; ontology; contextual targeting; audience.
Prakhar Rai, Indian Institute of Technology Guwahati, India
Natural Language Processing (NLP) for Big Data has become one of the most challenging frontiers in computer science and engineering. The exponential growth of heterogeneous, multi-modal, and noisy data has pushed NLP beyond classical statistical methods into the realms of distributed deep learning, knowledge-enhanced reasoning, and quantum-inspired architectures. This paper critically analyzes the integration of NLP with Big Data ecosystems, highlighting the interplay of scalability, semantic representation, and distributed optimization. We propose an advanced taxonomy of techniques, comparative analyses of architectures, and future research directions such as neurosymbolic fusion, federated multi-lingual embeddings, and quantum variational NLP models. Our aim is to stimulate the development of next-generation NLP systems capable of thriving in petabyte- scale, privacy-aware, and real-time environments..
NLP for Big Data, Deep Learning, Semantic Systems, Quantum NLP, Dis- tributed Architectures.
Satsuki Maeda, Bismark Kweku Asiedu Asante, and Hiroki Imamura, Soka University of Japan, Japan
Action recognition has many practical applications, but the task still faces significant challenges. A major challenge is the variation in human action poses across different viewpoints, which complicates determining the ideal pose for action recognition. To address these viewpoint-invariant issues, we propose a pose normalization approach combined with object-based action recognition to classify actions in videos. In this method, the normalized pose is compared with a reference pose to identify the action being performed. The objective of this research is to develop a three-dimensional (3D) object-associated action recognition framework that leverages the stereo camera’s ability to capture accurate distance information. This approach offers three main advantages: (1) action recognition that incorporates object context, (2) resolving occlusion problems, and (3) improving recognition accuracy through precise distance information. Experimental results show that our proposed approach achieves 70% classification accuracy across ten selected action categories, independent of viewpoint or camera angle.
Object Recognition, Behavior Recognition, AI, Stereo Camera, Active Detection.
Takaki Ohoka, Bismark Kweku Asiedu Asante and Hiroki Imamura, Soka University of Japan, Japan
In the interior and exterior industry, professionals goes through a painstaking tasks to plan and create functional and aesthetically pleasing indoor and outdoor spaces by selecting materials, colors, finishes, furnitures, and landscaping elements to achieve a harmonious and appealing environment for occupants and visitors. Recently, technological advancements has provided simulation environmemts using Virtual Reality (VR) or Augment Reality (AR) to easily perform these tasks. However the challenge of seamlessly switching between interior and exterior environments while simulating in multi user platforms is still a challenge. This is as a result of each of the spaces are simulated differently in VR space. To address this challenge, our research focus on developing a seamless simulation system for interior and exterior design in a VR space. Our system demonstrate that users can easily switch between interior and exteriors designs in the VR spaces in the multi user platforms as well.
Virtual Reality, interior design, exterior design, seamless simulations. 3D modelling, Data Visualization
Prakhar Rai, Indian Institute Of Technology Guwahati
The convergence of 3D imaging, biomedical applications, and secure data transmission presents a formidable challenge at the intersection of digital image processing and pattern recognition. This paper proposes a novel, end-to-end hybrid framework for the secure acquisition, processing, and transmission of 3D biomedical reconstructions. The core of our method lies in a unique combination of several advanced themes: we first employ a multidimensional signal processing pipeline for 3D and surface reconstruction from multi-view stereo image acquisition systems, addressing the critical issue of motion detection and illumination and reflectance modeling in clinical environments. The reconstructed 3D model then undergoes a dual-process: (i) An adaptive neuro-filtering stage, implemented on an embedded system (DSP Implementation), performs real-time array signal processing and higher-order spectral analysis to denoise and enhance the model, a task particularly crucial for low-signal medical image analysis. Concurrently, (ii) a deep learning-based watermarking algorithm, designed within a constraint processing framework, intricately embeds patient metadata and authentication signals directly into the spectral coefficients of the 3D mesh, ensuring data integrity and security without compromising visual fidelity for computer vision & VR-based visualization. This nonlinear approach to simultaneous enhancement and encryption is a significant departure from sequential methods. Our results, validated on a novel biomedical imaging dataset of anatomical phantoms, demonstrate a 23% improvement in reconstruction accuracy (PSNR) over state-of-the-art structure from motion techniques, a 40% reduction in signal noise, and unparalleled robustness of the watermarked data against common signal processing attacks, thereby establishing a new benchmark for secure, high-fidelity 3D telemedicine applications.
3D Surface Reconstruction, Adaptive Neuro-Filtering, Deep Watermarking, Multidimen sional Signal Processing, Biomedical Imaging, Embedded DSP, Secure Telemedicine
Ye Li, University of Canberra, Australia
Phishing persists because current email architectures cannot guarantee individual-level sender authenticity. Domain-based defenses (SPF, DKIM, DMARC) validate organizational infrastructure yet admit execution paths where messages from compromised accounts are accepted without verifying the named individual. We introduce authenticity completeness, a global property requiring that every execution of the email transmission process terminate either in verified authenticity or explicit rejection, thereby forbidding any form of unverified acceptance. We formalize this property via a state-machine model, prove that domain-level mechanisms are structurally incapable of satisfying it, and present a document-oriented digital signature scheme that uses ephemeral per-message keys. The design satisfies four reinforcing properties—proactive defense, individual-level authenticity, transparency of operation, and localized deploy ability—and we prove that it achieves authenticity completeness. This yields the first rigorous end-to-end guarantee against individual level impersonation within email systems.
Email Authenticity, Phishing, Digital Signatures, Authenticity Completeness.
Milind Jagre, Jia Huang, Dayvid V. R. Oliveira, Zhinan Cheng, Babak Seyed Aghazadeh, Puja Das, Chris Alvino, Jinda Han, Kailash Thiyagarajan, USA
In search and recommendation systems, predictive models often suffer from temporal instability when certain features introduce volatility in output scores. This instability reduces reliability and user experience - especially in multi-stage systems where consistent predictions are critical. We introduce Fortress, a general framework that enhances model stability and accuracy by identifying and pruning features that cause inconsistent scores over time. Fortress leverages temporally partitioned historical snapshots to capture score fluctuations for the same entity and follows a four-step process: (1) collect snapshots, (2) detect unstable predictions, (3) isolate instability-inducing features, and (4) retrain models with stable features. While semantic features from LLMs improve generalization, they often lack full coverage; engagement features add predictive power but introduce volatility. Fortress suppresses this instability while preserving value, yielding more stable and accurate models. Validated on a query-to-app relevance model in a large marketplace, Fortress shows notable gains in stability and PR-AUC.
recommendation system, relevance stability, feature pruning, information retrieval .
Ala Karajeh1 and Rasit Eskicioglu2, 1Independent Researcher, Winnipeg, Canada, 2Department of Data Science and Analytics, Atlas University, Turkey
Older patients typically exhibit different traits compared to younger ones and often have multiple comorbidities, such as diabetes and cardiovascular diseases, which complicate their severity assessment at emergency care facilities. This research utilizes two clinical databases from Beth Israel Deaconess Medical Center to explore the clinical characteristics of this demographic based on triage information, triage scores, and disposition outcomes. Additionally, a machine-learning model is proposed to predict likely disposition outcomes, specifically whether patients are hospitalized or discharged at the end of their emergency visit. This model could be instrumental in proactively managing this critical patient segment and improving their health outcomes.
Emergency Medicine, Hospitalization Prediction by Machine Learning, Emergency Older Patients Classification, and Emergency Older Patients Data Analytics.
Zhansheng Huang1, Garret Washburn2, 1USA, 2California State Polytechnic University, Pomona, CA 91768
Children face dual risks during school activities: safety concerns from inadequate supervision and health risks from carrying heavy backpacks exceeding recommended weight limits. BackTracked addresses both through an integrated IoT system combining GPS location tracking with real-time weight monitoring [9]. The solution comprises Particle Boron firmware interfacing with GPS and pressure sensors, a cloud backend for data processing and secure storage, and a cross-platform Flutter mobile application for parental monitoring [10]. Key technical components include NMEA sentence parsing for location extraction, analog pressure sensing with calibrated weight mapping, and token-based authentication for data security. Experimental evaluation demonstrated 3.2 meters outdoor GPS accuracy and 2.98% weight measurement error, validating system reliability for intended use cases. Compared to existing solutions, BackTracked uniquely combines health and safety monitoring, provides cellular connectivity independent of Wi-Fi, and supports both iOS and Android platforms. The system enables parents to monitor their childrens location and backpack weight through a single, comprehensive solution.
IoT Monitoring, GPS Tracking, Weight Sensing, Child Safety Systems.
Jamilynn Mackenzie Modelo and Mervis Encelan, College of Information Systems and Technology Management, Information Technology Department, Pamantasan ng Lungsod ng Maynila, Manila, Philippines
Menstrual health plays a critical role in an individual’s overall well-being, influencing physical, emotional, and reproductive health. Tracking menstrual cycles provides valuable insights into hormonal patterns, allowing for early detection of disorders such as hormonal imbalances. However, most period tracking applications are designed for users with regular cycles, often resulting in inaccurate predictions for individuals with irregular menstruation or Polycystic Ovary Syndrome (PCOS). This study proposes the development of Evelune, a menstrual and PCOS management mobile application that integrates KMeans Clustering and a rule-based prediction system to improve menstrual phase prediction and health awareness. The K-Means algorithm organizes user input into groups based on similarities in symptom patterns, which are then analyzed through a rule-based system that predicts menstrual phases and provides personalized health insights. Evelune further implements data privacy features such as AES-256 encryption, Firebase Authentication, HTTPS communication, and Multi-Factor Authentication (MFA) to ensure user data protection. The system is evaluated using ISO/IEC 25010:2023 quality standards, emphasizing functional suitability, usability, reliability, and security. Evelune aims to empower Filipino users, particularly those managing PCOS, by promoting reproductive health literacy, accurate symptom tracking, and privacy-conscious self-care practices
Menstrual Health, Polycystic Ovary Syndrome (PCOS), K-Means Clustering, Rule-Based System, Mobile Health Application, Data Application, Data Privacy, Reproductive Health Literacy.
Copyright © ACITY 2025
