This article aims to contribute as a preliminary research effort, providing a justification framework for the configuration of an Internet of Things (IoT) network designed to monitor the nonverbal communication behaviour of soldiers during training and mental preparation for the battlefield. Based on the results of this research, the network architecture of the IoT system will be configured in order to support the optimization of soldiers' nonverbal communication responses. Accordingly, the study focuses on analysing the role of two nonverbal indicators specific to the training and mental preparation process for the battlefield: physical proximity and paralinguistic aspects, which may reveal important information regarding levels of comfort, stress, concentration, or emotional control. By analysing these parameters, the paper proposes an approach for evaluating the effects of nonverbal communication on training performance and for identifying potential solutions to optimize remote training programs through an IoT network currently underAdevelopment.
Hand gesture recognition as the computer interpretation of human hand gestures used to assist people with disabilities. The texture-based feature descriptors of hand sign images do not have sufficient distinct features for recognition due to the change in geometric viewpoint and illumination invariant images. For instance, texture-based descriptors such as local binary patterns are sensitive to noise images. To overcome this problem, this paper proposes a novel global information extraction called contrast information fractal dimension (CIFD), which is based on contrast entropy and fractal dimension. The proposed CIFD method is used to obtain distinct features by analysing the edge quantity of hand sign images through measuring the information content, using non-linear filtering techniques and contrast entropy based on Weber and Devries-Rose contrast measurement. The extracted information from the hand edge images is applied to the convolutional neural network (CNN) for better hand sign recognition. The CNN is used to learn complex and non-linear relationships in static hand gesture images for recognition. The efficiency of the proposed system is verified by comparing various texture-based feature descriptors using standard databases. The hand sign recognition accuracy of the Jochen Triesch database for uniform and dark background environments are 99.50% and 95% and that of the National University of Singapore database (NUS-I) for black-white and colour background environments are 92.50% and 95% respectively.
This study explores how smart city strategies can improve access to cultural and entertainment events in Bucharest for young people aged 18-25, with a focus on marginalized groups. Using a mixedmethods approach, the research combines quantitative and qualitative data to find insight into barriers and user needs. Results reveal that 73% of the participants are willing to use an app with great interest in free utilization, a simple interface and geolocation-based prediction for available parking space. These ideas formed Forfest, a Next.js-based Progressive Web Application built with JavaScript (JS) and Tailwind CSS features, including offline browsing, a responsive design approach, multilanguage support and locationbased filters. The study illustrates the importance of user-centered tools for promoting cultural inclusion and digital citizenship.
The necessity for trustworthy Software Defect Prediction (SDP) models is highlighted by the increasing complexity of contemporary software systems. These models facilitate the effective use of scarce testing resources by early detection of potentially defective modules. Although deep learning has demonstrated promise in learning characteristics from source code, the efficacy of current methods is generally limited by their reliance on a particular sort of information, such as hand-crafted code metrics or semantic features from code structure. One of the biggest challenges is still integrating several data types into a single, discriminative feature set. In order to forecast file-level defects, this study presents a unique approach that blends semantic characteristics with source code metrics. In addition to Combined Defect Data Modelling (CDDM) we suggest Learning Hybrid Feature Representation (LHFR), a deep neural network model. LHFR combines a Multi-Layer Perceptron (MLP) to learn from manually constructed metrics with a Bidirectional Long Short-Term Memory (Bi-LSTM) network to extract semantic features from Abstract Syntax Trees (ASTs). With an average F-measure of 69.08%, LHFR outperforms models based solely on metrics or semantic characteristics when tested on 12 open-source Java projects. A new combined dataset, an improved feature set and a hybrid representation strategy that significantly enhances fault detection performance are among the contributions.
Human-Robot Interaction (HRI) is currently undergoing a paradigm shift from discrete commandbased control to seamless, symbiotic communication. However, achieving this symbiosis requires overcoming significant challenges in temporal alignment and computational efficiency, particularly when processing conflicting multimodal signals. This paper presents a hybrid study combining a semi-systematic literature review (2020-2025) with a novel conceptual framework. The study critically evaluates the architectural evolution from standard Transformers to linear-complexity State Space Models (SSMs), such as Mamba, highlighting the trade-offs between long-term semantic context and real-time responsiveness. Furthermore, the Multimodal Perception-Driven Decision-Making (MPDDM) framework is introduced as a conceptual model designed to resolve conflicts between explicit commands and implicit physiological cues via a dynamic confidence-weighting mechanism. To address the limitations of traditional latency-based benchmarks, a new set of evaluation metrics, specifically Modality-Specific Fluidity (MSF), is proposed to quantify the smoothness of interaction. Finally, recent integration paradigms are categorized into modular and end-to-end Vision-Language-Action (VLA) models, offering a critical synthesis of their respective safety and efficiency profiles. This work provides a roadmap for developing verifiable, low-latency HRI systems capable of operating in dynamic, unstructured environments.
Modern agriculture is significantly influenced by Machine Learning (ML), Deep Learning (DL), Internet of Things (IoT), and Cloud Computing. Particularly in soil health monitoring, crop assessment, nutrient management and surveillance. This paper analyses 164 peer-reviewed research papers published from 2011-2025 with the aim of examining advanced methodologies, performance trends, and limitations in AI-based agricultural systems. The research insights reveal that ML and DL models demonstrated high accuracy in isolated tasks such as prediction of soil parameters, crop disease detection and identification of nutrient deficiency. The existing solutions remain fragmented, cloud- centric, and inaccessible to resourceconstrained environments such as agriculture fields of smallholder farmers, especially with respect to wildlife intrusion deterrence. Based on these research gaps, this paper proposes a conceptual Scarecrow AgriBot framework that integrates edge-enabled, intelligent deterrence soil, crop monitoring, nutrient advisory and multilingual farmer interactions within a unified system. This Scarecrow AgriBot is a conceptual framework that serves as a literature-informed foundation for future prototype development and field-level evaluation.
One of the most vital strategic commodities with particular attention and popularity is crude oil since it affects people's daily life and is used in several sectors of industry. Globally, political and economic events affect crude oil prices constantly. Aiming to avoid financial losses or guaranteed future profits, many oil-related companies study the market to forecast prices. This work forecasts future crude oil prices using time series approaches. The Autoregressive Integrated Moving Average (ARIMA), Seasonal Autoregressive Integrated Moving Average with eXogenous (SARIMAX), and Long-Short-Term Memory (LSTM) models are applied to a real dataset spanning 12,055 days of oil prices. The three algorithms are to be compared in order to find the most accurate model for a crude oil price projection. The root mean squared error (RMSE) and mean absolute percentage error (MAPE) help to assess accuracy. With an RMSE of 0.02 and a MAPE of 2.6% SARIMAX shows to be better than the other models.
The study proposes a technology for identifying specialists, capable of performing specific tasks within software development projects. Existing models for describing professional competencies and task requirements are analyzed, showing that current approaches do not allow a direct comparison between specialists and tasks. The proposed technology builds domain-specific term dictionaries that link each term to the document in which it appears. To avoid linguistic inconsistencies, all documents are automatically translated into English prior to the term extraction. Each document model stores identified terms, technologies, and author contributions. The specialist model links authors to documents, enabling the representation of the individual competencies through associated technologies and terms. The task model represents assignments as texts containing relevant terms and technologies. A comparison mechanism determines which specialist or group of specialists best meets the task requirements. The experimental validation using the Task Distribution2 software demonstrated a 3.1-fold reduction in the time spent on taskspecialist matching while maintaining accuracy. The results confirm the efficiency of the proposed decisionsupport technology for selecting experts in multi-project environments.
The article outlines how the cyber diplomacy is becoming increasingly pertinent in light of the digital platforms' growing role in the political discourse, particularly in the aftermath of fake news becoming a significant challenge. It highlights that while the majority of the research studies target the identification of fake news using Natural Language Processing (NLP) and Machine Learning (ML), very few consider the economic implications. The paper proposes an Artificial Intelligence (AI) Open-Source Intelligence (OSINT) system that includes an automated content collection, a fake news identification engine, and an economic value model. The system would calculate the economic potential created by the disinformation operations. A web application prototype demonstrates the feasibility with promising results in detection ability and economic value. The solution provided can help governments and organisations by providing the information that bridges the detection capabilities of technology with the economic valuation.
This study examines the impact of ESG news sentiment on Romanian stock prices from 2019 to 2023, highlighting gaps in how an emerging market processes ESG-related information. The authors collected ESG-related web content through automated web scraping followed by manual data validation. Sentiment analysis was conducted in Python using the VADER model, and the results were linked to historical stock prices using regression analysis. The analysis revealed a delayed market response, with no significant effect on day t+1, but significant changes on days t+2 and t+3. Positive ESG news was linked to gradual increases in stock returns and reductions in trading volumes. The findings indicate inefficiencies in ESG information processing, highlight potential arbitrage opportunities, and underscore the need for better ESG disclosures and greater investor engagement.
Alzheimer’s disease (AD) requires early detection and continuous monitoring to enable effective management. Digital health platforms increasingly support this need, yet many lack usability, personalization, or integration across patient, caregiver, and clinician roles. This paper presents the NeuroPredict platform, a modular, role-based digital architecture which includes validated clinical assessment tools specific to AD. In compliance with the user-centered design (UCD) principles, the system provides automated scoring, longitudinal monitoring, and improved usability by adapting access and interaction to the requirements of the patients, caregivers, and clinicians. Internal assessments examined the reliability of differentiated workflows, scoring the algorithm accuracy, and platform stability. When compared to the current approaches that digitize instruments independently, the NeuroPredict platform shows how a role-based approach may enhance interpretability, ensure continuity of care, and facilitate integration into clinical workflows. These findings establish a solid technical basis for digital ecosystems in Alzheimer's disease monitoring which are clinically relevant, extensible, and interoperable, even though the validation with patient cohorts remains a future step.
Amid the accelerating digital transition, the knowledge systems face growing risks of fragmentation and epistemic discontinuity. To address these challenges, the public libraries are increasingly recognized as strategic socio-technical infrastructures that sustain the recovery, renewal, and long-term preservation of knowledge. Building on the view of knowledge as a renewable resource - one that can be created, reinterpreted, and recombined across social contexts - this paper introduces the concept of renewable knowledge, defined as the dynamic property of knowledge to be continuously updated, co-created, and adapted through human-service-technology interaction for public value creation. Anchored in the Service Science and Knowledge Commons theory, the study examines how public libraries operate as aggregators of renewable knowledge through the open data mediation, big data valorisation, and AI-assisted information services. Employing a mixed-methods design (survey and structured interviews across Romanian and European public libraries), the empirical results identify six conceptual clusters reflecting how libraries enact renewable knowledge via creation, utilization, organization, and distribution. Findings demonstrate that libraries move beyond static information management toward participatory knowledge stewardship, fostering resilience, sustainability, and collective intelligence within socio-technical ecosystems.
Examining vast and complicated data sets to find patterns, correlations, and market shifts in order to improve business decisions and insights is known as Big Data Analytics. It entails the use of sophisticated methods and instruments to gather, process, and evaluate diverse, high-volume, and high-velocity information gathered from sources such as sensors, social media, and logs of transactions. Due to the rapid expansion of data and technological advancements, Big Data Analytics (BDA) has attracted a lot of attention from both academics and industry, especially in fields like healthcare. By increasing results for patients, enabling tailored medication, and boosting the precision of diagnostics, the combination of data from many sources and the application of cutting-edge analytical tools have the capacity to completely transform healthcare. Healthcare data, big data in healthcare organizations, and the uses and benefits of big data analytics in healthcare are all covered in the present research. It is also discussed how big data in healthcare has advanced technologically, including comparison of tools for Big Data Analytics. The discussion also includes the difficulties with big data analytics in healthcare settings.
The rapid growth of the video content across online platforms has made it increasingly important to generate concise summaries that help users quickly understand and navigate long videos. However, creating high-quality video summaries typically requires large amounts of annotated data, which is costly and often unavailable. To address this challenge, the authors propose a fully unsupervised approach to video summarization built on Transformer architectures. The method introduces the Reinforced Encoder-Decoder Summarizer Model (RED-SM), which uses multi-head self-attention and feature extraction to identify informative video segments without human labels. RED-SM incorporates sparsity-promoting penalties and a reinforcement learning reward that balances diversity, representativeness, and temporal smoothness to guide frame selection. To further enhance the summarization quality, the RED-SM with a BERT-based text extractor is integrated, enabling multimodal fusion of visual and textual cues. The approach is evaluated on the SumMe and TVSum datasets, as well as a newly curated dataset of 30 categories of short videos. The experiments show that the method consistently produces concise and high-quality summaries across diverse domains. These results highlight the RED-SM as an effective and scalable solution for unsupervised video summarization in real-world applications.
This study presents a comparative analysis of the effectiveness of single-objective algorithms in optimizing automated process control. The research is conducted on a flexible manufacturing system (FMS) comprising seven production stations, each equipped with an energy consumption monitoring system. The objective is to examine how the optimization algorithms can contribute to enhancing the operational efficiency of the production systems. The study evaluates several single-objective algorithms - Genetic Algorithms (GA), Particle Swarm Optimization (PSO), Gray Wolf Optimization (GWO), Ant Colony Optimization (ACO) to assess their potential for energy optimization in flexible manufacturing processes on FMS. Each algorithm's strengths and limitations are discussed with respect to their effectiveness in minimizing energy consumption and enhancing system performance. A comparative evaluation of the results obtained through the implementation and testing of each algorithm highlighted the superiority of the GWO algorithm.
In the digital era, the security, integrity, and authenticity of the digital forensic evidence are crucial in a judicial information system. In the paper a system that enables users to securely store the digital evidence used in legal proceedings within an architecture designed to withstand various types of attacks is proposed. It implements 12 security solutions that ensure the authenticity and protection of the stored data and prevent the interception of information within the digital evidence repository. The system integrates multiple solutions, including blockchain, data encryption techniques, communication encryption, customized solutions for immutable data storage, and access restriction through advanced filtering mechanisms. It also provides rolebased access control to the digital content, ensuring the strict permission management. The implementation results demonstrate that the system is secure, robust, and capable of handling large files efficiently.
Agile software development necessitates the definition of user stories; however, the conventional manual process frequently leads to inefficiencies and inconsistencies. This paper investigates the shift from the deterministic user story creation to the AI-powered automation, addressing the limitations of the manual methods and the benefits of the AI integration. The study highlights the development of a deterministic JIRA extension for e-commerce applications. Despite its efficiency, the deterministic tool lacked adaptability across different domains. To overcome this, an AI-powered enhancement utilizing OpenAI's API was introduced, enabling scalable and accurate user story generation through natural language processing and machine learning. AI-driven tools automate user story creation, improving accuracy, reducing misinterpretation risks, and streamlining workflows. The research traces the evolution from a deterministic automation app using templates to an AI-driven app, emphasizing the role of prompt engineering in refining AI-generated outputs. The results demonstrate that AI integration not only enhances efficiency but also extends user story definition to diverse application domains, contributing to more scalable and adaptable software development practices.
The future of the wireless networks holds a significant promise, particularly in the domain of the Mobile Ad hoc NETworks (MANETs), which continue to attract a growing academic interest. As MANET usage expands across diverse applications and user demands increase, an enhanced performance and reliability have become critical. Ensuring the Quality of Service (QoS) is essential for effective communication, with load balancing in routing protocols playing a vital role in optimizing the resource management and improving the overall network performance. In this study, two enhanced multipath routing protocols are proposed - SMMSN-AOMDV (Stable Multi-path Selection with More Stable Nodes in AOMDV) and MAN-AOMDV (Multi-path Routing with Available Nodes in AOMDV)-specifically designed to improve route stability and load distribution in MANETs. The approach introduces a novel load balancing mechanism within the AOMDV protocol by selecting the paths based on node-level parameters and dynamically distributing traffic across less congested nodes during data transmission. The proposed protocols are implemented and evaluated using NS2 (Network Simulator 2). The simulation results demonstrate that the enhanced protocols significantly improve the key performance metrics, including the packet delivery ratio, throughput, and end-to-end delay.
In this work, an advanced multimodal video summarization system is developed for a clustercontrolled traffic surveillance environment. Addressing the growing challenges of vehicle density and traffic complexity, the system employs state-of-the-art deep learning techniques for real-time vehicle detection, tracking, and route analysis. Utilizing algorithms like YOLO for object detection, the system ensures an accurate and efficient monitoring across multiple surveillance nodes. The optical character recognition (OCR) enables a detailed number plate recognition, while the multimodal data fusion enhances the robustness of vehicle tracking in dynamic conditions. Supported by data pipelines and frameworks, the system processes extensive CCTV footage to generate concise video summaries, optimizing the surveillance operations. The architecture offers scalable and adaptive solutions, aiming to improve the traffic management and emergency response by providing actionable insights through seamless integration of visual, audio, and textual data. This innovation has the potential to set new benchmarks in the intelligent traffic surveillance systems, with significant implications for public safety and urban mobility.
