Authors - Ubaydullo Vakhabovich Gafurov, Nargiza Ruzibaeva, Khujayar Musurmonovich Shennayev, Nosir Mahmudovich Mahmudov, Saodat Sadriddinova, Alisher Bakberganovich Sherov, Nilufar Karimova Abstract - The sustainable development of tourism enterprises increasingly depends on their capacity to implement innovative activities supported by efficient financial mechanisms. However, traditional financing models often fail to provide sufficient flexibility and accessibility for innovation-driven tourism firms, particularly small and medium-sized enterprises. This study proposes a digital financial ecosystem model designed to enhance the financing of innovative activities in tourism enterprises through digitalization instruments. The research integrates concepts of digital finance, FinTech platforms, data-driven credit assessment, and smart contract mechanisms into a unified systemic framework. Using a system dynamics modeling approach combined with structural analysis, the study develops a conceptual and quantitative model linking digital financial infrastructure, access to alternative funding sources, and innovation performance indicators. Empirical validation is conducted using survey data collected from tourism enterprises and analyzed through structural equation modeling. The results demonstrate that digital financial ecosystem components significantly improve funding accessibility, reduce transaction costs, and increase innovation intensity. The findings highlight the mediating role of digital maturity in strengthening the relationship between financial accessibility and innovation outcomes. The proposed model contributes to the theoretical development of digital transformation in tourism finance and offers practical implications for policymakers and enterprise managers. The study provides a scalable framework for enhancing innovation financing in digitally transforming tourism markets.
Authors - Heidi Koivisto, Sari Ahonen Abstract - The integration of AI-assisted code generation tools in software development has the potential to significantly improve productivity and code quality. This paper presents a comparative study of several leading AI tools, including GitHub Copilot, Amazon Q, GitLab Duo, and Claude Code, in the context of software quality assurance (QA). These tools are evaluated based on their ability to generate test cases for a sample application, focusing on metrics such as test coverage, accuracy, and maintainability. Our findings provide insights into the strengths and limitations of each tool, offering guidance to practitioners seeking to take advantage of AI in their QA processes. The results indicate that while AI tools can accelerate test case generation and improve coverage, careful consideration is needed to ensure the generated tests are relevant and maintainable in the long term.
Authors - Avanti Chokhare, Reena Satpute Abstract - With the world's population moving from rural to urban areas, urban infrastructure including, but not limited to, transportation systems, energy re-sources and public service delivery are all strained in their ability to cope with additional residents. Urban management techniques that have historically worked in managing urban growth are no longer sufficient to meet the demands of com-plex modern cities. The Internet of Things (IoT), a transformative technology that allows real-time data collection, enables smarter decision-making and automates many of the services within urban areas through a variety of connected sensors, devices and digital platforms. The purpose of this study is to analyze how IoT will enable the creation of sustainable and efficient smart cities by encompassing multiple transportation management systems, energy efficiency improvements, and environmental monitoring, health, education and digital governance administrative systems. Additionally, the study investigates additional enabling technologies such as Artificial Intelligence (AI), Edge and Cloud Computing, Block-chain and other technologies that can improve the overall functionality of the IoT. Finally, this paper investigates key challenges associated with the implementation of IoT applications in urban areas, including security, reliability and data privacy issues. It is clear that the implementation of standardized frameworks, scalable architecture and phased implementation strategies will be key components in the successful implementation of smart cities. Responsible adoption of IoT technology will lead to improved environmental sustainability, operational efficiency and citizen quality of life in urban areas.
Authors - Joseph Afriyie, Stephen Opoku Oppong, Benjamin Ghansah, Daniel Kobina Danso Essel, Dickson Keddy Wornyo, Ephrem Kwaa Aidoo Abstract - The COVID-19 pandemic forced educational institutions to adopt online learning, which resulted in expanded digital spaces that cybercriminals used to launch phishing attacks against students, faculty, and institutional systems. This research article provides a comprehen-sive literature review that evaluates machine learning techniques for phishing detection in online educational settings. The PRISMA guidelines were used to select 40 studies from 2013 to 2023 after researchers examined publications retrieved from IEEE Xplore, SpringerLink, and Google Scholar. The review analyzes various digital education ecosystems through its examination of algorithmic methods and datasets, performance evaluation metrics, and detection framework de-signs that universities use to defend against phishing attacks. The research shows that Random Forest and Gradient Boosting, together with deep learning methods, which include Convolutional Neural Networks, Long Short-Term Memory networks, and Recurrent Neural Networks, deliver superior detection performance reaching over 90% accuracy in most scenarios. Educational in-stitutions encounter three primary challenges, which include implementing real-time systems to combat emerging phishing techniques, ensuring dataset compatibility with various environments, and managing their restricted resource availability. The study establishes that institutions need to create technical detection frameworks that work together with user training programs to establish better institutional protection measures.
Authors - Bullet Tiwari, Reena Satput Abstract - The Internet of Things (IoT) systems are becoming numerous, trans-forming our virtual world by continually gathering information, linking equipment, and automating most of the spaces. However, the size, decentralization, and dispersal of the IoT networks cause grave droughts with reliability, security, and workability. The current rule-based surveillance tools are not able to handle the dynamism and volume of data generated by drastically many IoT devices. The paper describes a machine learning (ML) system that predicts the performance of applications, issues, and predicts failures of devices in IoT settings. It compares the methods of supervised, unsupervised, and deep learning and examines their performance in the constraints of computing power, delay and energy. It talks about the trade-offs between centralized and decentralized learning in which clouds and edges are used respectively to decide on the most suitable deployment. The framework has security, privacy, and sustainability design issues, as well. The suggested solution is expected to enhance IoT resiliency with the help of predictive intelligence to manage issues before they happen and increase the overall stability of the industry, healthcare, and smart city environments.
Authors - Roseline Oluwaseun Ogundokun, Rotimi-Williams Bello, Pius Adewale Owolawi, Chunling Tu, Etienne A. van Wyk Abstract - Extramarital affairs can undermine trust and lead to relational break-down, yet the ability to anticipate risk factors remains limited. Recent studies have used deep learning approaches to predict infidelity, achieving high classification accuracy but at the cost of interpretability and resource requirements. This article proposes a novel research to examine whether simple, interpretable mod-els, k-nearest neighbours (KNN), linear regression (LR) and support vector regression (SVR), can predict the amount of time individuals spend in extramarital affairs using readily available socio-demographic and relational features. Using the well-known affairs dataset comprising 6,366 observations and nine variables, we apply feature engineering, cross-validation training and regression-based evaluation to compare model performance. Our findings indicate that although KNN outperforms LR and SVR in terms of accuracy, all models struggle to capture variance; mean squared error (MSE) values remain high, and the coefficient of determination (𝑅2) values close to zero. We discuss the implications for predictive counselling and outline future research directions.
Authors - Manar Eloued , Narjes Benameur , Sonia Esseghaier, Salam Labidi Abstract - Magnetic Resonance Elastography (MRE) is a novel, non-invasive im-aging technique for assessing liver stiffness. However, the lack of standardization introduces variability in measurements. The Manual selection of the Region of Interest (ROI) remains subjective and operator-dependent, often including areas with blood vessels or poor wave propagation, which can compromise measurement accuracy. This study proposes a deep learning- based approach to automatically identify an optimal region for liver stiffness measurement (LSM). A total of 160 MRE ex-ams, comprising paired magnitude and wave attenuation images from both healthy individuals and patients with liver disease, were used. A 3D U-Net architecture was trained to segment the liver, blood vessels, gallbladder, and biliary ducts from magnitude images, as well as regions of good wave propagation from attenuation images. The final ROI was obtained by intersecting these segmented regions. The model performance was evaluated on a separate test set using the Dice Similarity Coefficient (DSC), paired Student’s t-test, and Bland-Altman analysis. The resulting LSM region achieved a DSC of 0.89. The t-test yielded p = 0.68, indicating no significant difference between the automated and manual ROIs (p > 0.05). This automated pipeline reduced analysis time from approximately 20 minutes manually to less than 10 seconds automatically, while ensuring reproducibility and reducing operator dependency in MRE by standardizing ROI selection while maintaining diagnostic accuracy. It offers a promising solution to improve the reliability of LSM, particularly for longitudinal follow-up.
Authors - Hector Rafael Morano Okuno Abstract - One of the applications of LLMs (Large Language Models) has been their use as assistants, enabling users to perform specific tasks via prompts, from solving mathematical problems to generating images or videos. This article aims to explore the capabilities of an LLM in generating scripts for the API (Application Programming Interface) of the CAD software Fusion 360, identify the types of geometries it can create, and determine whether it can reproduce images in CAD models. This work was developed during the Manufacturing Systems Automation course for Mechatronics Engineering students, with the intention of introducing them to the use of LLMs in their field. Among the find-ings, it was determined that the user must be familiar with the Fusion 360 API to correct potential errors in the scripts generated by the LLM. Furthermore, the user must be able to specify, via prompts, the characteristics of the parts to be modeled, ensuring that the specifications are compatible with the instructions Fusion 360 understands. Regarding students' experience with LLMs, they found them useful, as they saved time in designing components that require complex automation systems.
Authors - Zilola Mamatvaliyevna Aliyeva, Nigora Primova, Dildor Abduraxmanovna Shadibekova, Malika Akbarova, Azizbek Mahkamov, Gulchehra Raxmatjonovna Xusanova, Shoh-Jakhon Khamdаmov Abstract - The construction industry plays a strategic role in Uzbekistan’s economic development; however, it continues to face challenges related to cost overruns, project delays, and limited financial transparency. Digital transformation offers new opportunities to enhance economic efficiency and project management performance. This study examines the impact of digitalization on construction economics in Uzbekistan, focusing on the implementation of Building Information Modeling (BIM), digital cost estimation systems, electronic procurement platforms, and enterprise resource planning (ERP) solutions. The research develops a conceptual model linking digital adoption level, cost control effectiveness, project performance, and financial outcomes. A quantitative survey of construction companies operating in Uzbekistan was conducted, and Structural Equation Modeling (SEM) was applied to test the proposed relationships. The findings indicate that higher levels of digital integration significantly improve cost estimation accuracy, reduce budget deviations, and shorten project completion time. Digital procurement systems also enhance financial transparency and reduce operational inefficiencies. The study provides empirical evidence that digital transformation positively influences economic performance in Uzbekistan’s construction sector. The results contribute to construction economics literature and offer policy recommendations for accelerating digital adoption in emerging markets.
Authors - Jorge Duque, Antonio Godinho, Jose Moreira, Firmino Silva Abstract - Student dropout in higher education remains a persistent academic, institutional and social challenge, requiring evidence-informed responses. This paper develops and evaluates an explainable machine learning artefact for early dropout-risk identification and for translating predictions into tiered institutional interventions. The study follows six phases of Design Science Research and uses the public Predict Students' Dropout and Academic Success benchmark, with 4,424 students and 37 variables. The pipeline integrates one-hot encoding, derived features, stratified validation, SMOTE applied only inside training folds, Random Forest, XGBoost and SVM as base learners, stacking with a logistic meta-learner and SHAP explanations. The final ensemble achieved 96.4% accuracy, 0.965 weighted precision, 0.964 weighted recall, 0.964 weighted F1-score and 0.99 weighted AUC. The contribution lies in combining performance, interpretability, governance and responsible human intervention.
Authors - Hasna Noushad, Geetha KN Abstract - The brain tumor remains a serious health concern and diagnosis in the primary stage is mandatory for effective treatment. Medical image analysis plays a vital role in understanding the underlying disturbances, monitoring, treatment planning, and intervention strategies. Graph theory is one of the popular techniques used for medical image analysis. The study focuses on the introduction of a systematic approach by using the application of vertex addition of graph theory to synthetically construct a glioma brain network utilizing the normal and abnormal brain Magnetic Resonance Image (MRI). The paper presents the idea of the initial emergence and growth of tumor from a graphtheoretical perspective. Comparative analyses are carried out between normal and abnormal brain network due to the growth of glioma. Results demonstrate that the presence of strong structural deformations and alterations in brain network due to the introduction and growth of glioma. In addition, the article also examines the corresponding increase in the correlation values of the tumor with other normal regions of brain as the tumor grows. This work provides a robust foundation for future studies in epidemiological modeling, machine learning and deep learning methodologies where lack of required data is an issue.
Authors - A Aruna kumari, Sri Vishnu Prabhu Gudavalli, Tamminana Visweswari Abstract - Diabetic Retinopathy (DR) is one of the biggest contributors of visual impairment and blindness in diabetic patients who do not receive proper measures and treatment at the early stages. Due to the ever rising instances of diabetes by the day, more concern has been raised on the effectiveness of methods of effective, scalable and early diagnosis. The given paper is a proposal of a deep learning-based algorithm of DR diagnosis, specifically, the algorithm named Patch-Wise Segmentation and Classification using Convolutional Neural Networks (CNNs). The system in this case is in contrast to the traditional systems, which require the entire retina image to be processed simultaneously by the system, whereby high-resolution fundus photographs are broken into patches. The model addresses each patch individually in order to have the model closely observe minute-scale details and sensitive pathological changes such as hemorrhages and exudates. Patch-wise processing dramatically enhances the capacity of reporting early and mild cases of DR that are hard to recognize in the fullimage processing due to intricacy of an image and noise. The CNN architecture is additionally medical image particular and operates by use of layers and regularization methods so as to permit not only accuracy but also generalization over a wide range of datasets. As shown in the results of the experiments, the patchwise technique performs better in comparison with the traditional ones, i.e., sensitivity, specificity, and classification accuracy on each of the stages of the DR. In addition, the system is fully automated and stable that minimizes the use of human marking and offers the facility to operate worldwide
Authors - Gavin Singh Pandha, Umar Khokhar, Binh Tran Abstract - This research initiative investigates the growing role of artificial intelligence (AI) in the automation of cyber-attacks and its emerging impact on modern cybersecurity. This study explores how AI models can be used to per-form web application intrusions by simulating attacks against deliberately vulnerable applications. Performance metrics from these simulations are compared with traditional, manually executed intrusion techniques, in addition to examining real world cases for AI misuse, ethical concern, industry standard, and the growing risk of autonomous threat actors
Authors - M. V. Rama Sundari, Bhuvan Unhelkar, Pravin Kshirsagar, Supriya Nandikolla Abstract - The management of nutrient content in soil is vital for improving productivity of crops, as well as sustainable agriculture. Conventional processes of establishing the best ratios of macronutrients tend to overlook intrinsic relation-ships that exist among soil properties, crop, weather. Main purpose is to come up with a smart prediction system that will help forecast Macronutrient needs by infusing particular domain-specific expertise in agriculture into machine learning models to improve interpretability and predictive accuracy. The suggested methodology utilises a Graph Convolutional Network (GCN) to simulate spatial and relational relationships amongst soil, crop and environmental parameters. To provide the model with a better semantic understanding, a Knowledge Graph (KG) is created to encode the relationships between domains. Embedding algorithms, such as TransE and DistMult, are then added to the GCN to form a KG-embedded GCN model that can learn feature-based as well as semantic relationships to predict nutrients. The experimental analyses on the ICFA Crop Recommendation dataset reveal that the baseline GCN got the R² scores of 0.806, 0.729, and 0.768, and the TransE GCN and DistMult GCN models have been advanced to the R² scores of 0.808-0.821, 0.746-0.762, and 0.800-0.814 for Nitrogen, Phosphorus and Potassium respectively. These findings indicate that the predictive strength is greatly advanced by the incorporation of domain knowledge. The model can, however, perform differently on unknown crop varieties and soils, which suggests that more work needs to be done in the future on larger and region-specific datasets.
Authors - Antonio Cortes Castillo Abstract - The rapid advancement of Artificial Intelligence (AI) has fundamentally transformed data center operations. The widespread adoption of AI services has introduced new requirements for hosting AI systems, prompting significant modifications in the design and construction of modern data centers. As a result, data center operators must implement comprehensive strategies to address the challenges associated with AI integration. This study explores the application of machine learning techniques using neural networks and Multilayer Perceptron (MLP) models for data center optimization. The research focuses on critical metrics, including power consumption, liquid cooling, fiber-optic systems, and the number of fiber-optic routing paths, which pose significant challenges for data center operators. Experimental results are analyzed using simulation tools, including SPSS, to demonstrate enhancements in data center performance and efficiency.
Authors - Hera Khairunnisa, Naomi Helena Elizabeth, Dwi Kismayanti Respati, Ayatulloh Michael Musyaffi, Gentiga Muhammad Zairin Abstract - This study presents a bibliometric analysis of the literature on accounting and non-profit organizations (NPOs) based on 248 documents retrieved from the Scopus database. Data were processed and visualized using Biblioshiny (R Studio) and Scopus web analysis. The findings indicate a significant growth in publications since 2000, peaking in 2020–2024. Accounting, Auditing and Accountability Journal emerged as the most dominant source, and the distribution of journals is consistent with Bradford's Law. The United States and United Kingdom lead in scientific contributions, while Indonesia shows a growing presence. Keyword analysis reveals a shift toward contemporary themes such as blockchain, sustainability, and social accounting. This study provides a systematic mapping of the intellectual landscape of NPO accounting research and identifies opportunities for future investigation.
Authors - Himangshu Sarma, Madhumita Banerjee, Chandrajit Choudhury Abstract - Diabetic Retinopathy starts at a light level with no visible symptoms, but it can lead to severe pain and blindness as the disease progresses. Clinically, DR is diagnosed by looking for retinal detachment or utilizing imaging techniques like fundus imaging or optical tomography. The Early Diabetic Retinopathy Study is one of the established DR staging schemes. Image Processing has played a significant role in improving the methods used to detect the disease automatically. It has a huge role in assisting ophthalmologists in the screening process, as manually screening by ophthalmologists consumes more time, sometimes there may also be error. Although there are a number of algorithms used in detection, therefore, in this work we have studied and implemented various DR detection methods. To differentiate among various classes, we also prepared KAGGLE APTOS dataset containing the GLCM, LTP, GLRLM, LMeP, CLBP, CSLBP and LBP features extracted from the image dataset. And the classification is performed in the dataset, whilst achieving an accuracy of 97% on testing dataset and 96% for training dataset using SVM multi classifier.
Authors - Najera R. Umpar, Minsoware S. Bacolod Abstract - In this study, the readiness of teachers in adopting Artificial Intelligence (AI) in teaching and learning, and the variables that influence their acceptance or resistance towards it, were investigated. Following a qualitative research approach, semi-structured interviews were employed to gather the data. It was found that a range of factors including generation of teachers, teaching discipline or specialization, institutional support, and ethical issues influence teachers' readiness in implementing AI in the teaching process. Young teachers, along with those who are specialized in STEM subject area, expressed confidence, preparedness, and enthusiasm in embracing AI in teaching. Experienced teachers, and teachers who taught other subject area, expressed concerns toward relevance and teachers' autonomy. Institutional support (were considered as a factor that would significantly impact teachers' readiness toward AI integration. Ethical concern such as student privacy, bias algorithm, and student monitoring have also contributed to teachers' beliefs on AI implementation. More important, the study identified "hybrid readiness" where teachers believe AI can serve as the "co-teacher" of them and contribute to individual learning and pedagogical practice. It suggests that there are variety of factors influencing the teachers' readiness toward AI implementation thus it requires more comprehensive planning in order to have more efficient integration of AI in the classroom. It is found that teachers' readiness is still not homogeneous and context dependent, therefore differentiated training and education as well as support policy are necessary to enhance teacher readiness and promote the integration of AI in education.
Authors - Kalinka Kaloyanova, Elitza Kaloyanova Abstract - Despite the increasing use of artificial intelligence (AI) in healthcare, clinician adoption of AI tools is still obstructed by algorithmic aversion, which reflects scepticism about the results of AI use. This article examines how hospitals can enhance AI adoption by strengthening AI competencies in physicians and mitigating mistrust through a systematic, data-driven approach. A review of behavioral studies reveals barriers to AI adoption across technical, cognitive, organizational, and ethical domains. A framework is proposed that focuses on integrating individual clinician competencies with structured strategies implemented by hospitals that support workflow and create conditions for continuous learning. The implementation of data-centric strategies, explainable AI tools, competency programs, simulation training, and interprofessional collaboration is recommended to increase trust in AI in medicine, support its ethical use, which ultimately leads to safer healthcare delivery.
Authors - Md Manirul Islam, Umme Salsabil, Md. Mushfiqur Rahman, Sazzad Hossain Abstract - This paper presents a compact identity-verification architecture for private web and Internet of Things (IoT) deployments that require tamper evidence without the operational overhead of a full blockchain. The framework separates credential verification from profile-integrity verification across multiple stores: a credential store, a protected-profile store, a reference integrity store, and a key store. Credentials are protected with Argon2id-based verifiers, while protected profile records are bound to entity identifiers, timestamps, and version counters through HMAC-SHA-256 reference tags. Unlike scan-heavy hash-only workflows, the proposed design performs direct indexed lookup by entity identifier and then verifies integrity through a keyed comparison step, improving both security posture and scalability. The same logic can be deployed behind HTTPSbased web services and MQTT-over-TLS IoT gateways. A reference prototype and benchmark study over datasets of 1,000 to 10,000 entities show that the indexed login path remains nearly size-stable, with median successful login latency around 1.68-1.69 ms under a development-profile Argon2id configuration, while a scan-based baseline login path grows from 0.92 ms to 6.90 ms over the same range. Injected profile tampering was detected in all benchmarked trials. The resulting framework offers a pragmatic middle path between conventional centralized login and heavyweight distributed-ledger authorization for institutions that prioritize local autonomy, compartmentalization, and data-integrity assurance.
Authors - MD Junayed Talukdar, Khosro Salmani Abstract - Artificial Intelligence (AI) systems are widespread across fields such as healthcare, finance, employment, and criminal justice, with a substantial impact on the lives of individuals and society. However, AI systems have been shown to perpetuate existing social inequalities, particularly through biases that are not easily discernible. Such biases are embedded in the technical and social structures of AI systems, posing a direct challenge to the principles of Equity, Diversity, and Inclusion (EDI) understood as the acknowledgment of differences among individuals, fairness and equal access, and the valuation of all participants. This study argues that fairness in AI cannot be achieved by focusing solely on technical aspects, necessitating a holistic approach. To investigate this, computational content analysis was applied to 360 occupational narratives generated by ChatGPT across nine professions and four geographic regions (Canada, Germany, India, and Bangladesh) using explicitly gender-neutral prompts. The analysis examined whether AI-generated narratives associate professions predominantly with one gender, and whether such patterns remain consistent across regions. Findings reveal that gender bias persists despite neutral prompting, with male-coded protagonists dominant in technical and manual labor professions and female-coded protagonists dominant in caregiving roles. Although regional conditions influenced the magnitude of gender imbalance, the direction of occupational gender patterns remained largely consistent across all four regions. This study identifies the empirical foundations necessary for future EDI-AI co-design frameworks, outlining the sociotechnical dimensions that such frameworks must address to be effective.
Authors - Yassine Lkhalidi, Mohamed Lkhalidi, Hatim Kharraz Aroussi, Achraf Tifernine Abstract - IoT device authentication remains vulnerable to credential theft and physical-layer impersonation, particularly where resource constraints preclude full PKI deployments. Existing approaches address subsets of this problem: RF fingerprinting exposes templates in plaintext, while zero-knowledge proof (ZKP) schemes authenticate static keys without binding to physical hardware. We propose ZK-RFAuth, a framework integrating Siamese CNN-based RF fingerprinting, Groth16 ZKP embedding verification, and Proof-of-Authority blockchain logging. A device’s hardware imperfections are captured as a compact embedding; a Groth16 circuit proves the L1 distance between a fresh embedding and the registered template falls below a predefined threshold, without revealing either vector. Evaluated on WiSig (28 WiFi transmitters, 224,000 I/Q frames), ZK-RFAuth achieves 91.4% closed-set accuracy, 2.25% Equal Error Rate, and 70.8% rogue rejection at the 95th-percentile operating threshold, requiring only 972 R1CS constraints for 144-byte proofs verified in approximately 3 milliseconds. ZK-RFAuth is the first framework providing physical-layer identity, embedding-level zero-knowledge privacy, open-set rogue detection, and immutable audit logging simultaneously.
Authors - Aryan Sharma, Dipali Baviskar Abstract - Modern computer networks are often equipped with an intrusion detection system (IDS) to detect malicious activities or cyber-attacks. Such a system must have high accuracy on known attacks, and at the same time it must be able to generalise to previously unseen attacks. However, supervised classifiers fail to generalise to new situations because they learn to map input data to output labels under a specific training distribution, and they perform poorly under a different test distribution, which is called distributional shift. In this paper, we propose a gating-based hybrid IDS that combines supervised classifiers with anomaly detectors. The gating network restricts the influence of the anomaly component to the uncertain prediction zone, i.e., the region of the output space where the classifier is uncertain, defined by a probability range of (0.15, 0.75)], and prevents unsupervised noise from affecting the confident supervised decisions. We evaluate the performance of our proposed system on three different test scenarios using the CIC-IDS-2017 dataset. The first test scenario consists of eight known attacks for which we train the classifiers on the corresponding training data, and then we test them on the corresponding test data. The second test scenario is an out-of-distribution stress test, in which we use 99% benign traffic and add DDoS and PortScan attacks to it, and test whether the system is able to detect them. The third test scenario is a zero-day test scenario in which we test the system on a previously unseen SQL Injection attack. Our findings are as follows. First, the Gating Hybrid RF+AE achieves an F1-score of 0.9778 and a precision of 0.9924 on the eight known attacks, which outperforms the standalone RF classifier with an F1-score of 0.9750. Secondly, on the out-of-distribution test scenario, both the RF and GBT classifiers fail to detect the attacks with an F1-score of 0.000, while the Gating Hybrid RF+IF achieves an F1-score of 0.405, which corresponds to a 40.5 percentage-point lift from the F1-score of the anomaly component IF. Thirdly, the Gating Hybrid RF+IF achieves an SQL Injection recall of 47.6% on the zero-day test scenario, while the standalone RF and GBT classifiers achieve an SQL Injection recall of 33.3% on average. All the abovementioned results are supported by 95% Wilson confidence intervals, and we provide root-cause analysis for the extreme results.
Authors - Moises Toapanta T, Jeanette Jordan Buenano, Nancy Jordan Buenano, Maria Cristina Espin Melendez, Pamela Toapanta Pavon, Rocio Llumiquinga A., Dafna Guaman B., Eriannys Gomez D., Pedro Echeverria B. Abstract - The globalization of Information and Communications Technologies (ICTs) the Internet, artificial intelligence (AI), and social media poses serious threats to the integrity, confidentiality, and authenticity of information in higher education institutions (HEIs). The central problem lies in the absence of robust legal frameworks regulating the use of AI, particularly in relation to personal data protection. This study examines perspectives on AI and social media, together with the legal foundations required for the effective administration of HEIs. Using the deductive method and exploratory research, key actors in institutional governance were identified, administrative strengthening indicators were de-fined, and an integrated model was developed to link AI, social media, and regulatory frameworks. It is concluded that improving institutional governance re-quires mitigating the risks associated with the use of these technologies through legal frameworks aligned with national constitutions and regulations. Ecuador, like most Latin American countries, currently lacks such legislation and remains in the analysis phase.
Authors - Bhavana Chalkalwar, Reena S. Satpute Abstract - In today’s world Artificial Intelligence is rapidly growing in modern tech industry which enhances the power of new IT services. Artificial Intelligence helps to Education, Government And Private Services for better future work. The new way of teaching and providing instruction requires an adaptive approach rather than the traditional way of teaching to meet the wide variety of student learners. This change in education is being led by technological advances in areas such as machine learning, natural language processing, and intelligent tutoring systems. This paper will examine Artificial Intelligence's role in education, including the various applications available and what potential benefits and challenges each might present, as well as how the use of Artificial Intelligence in education as a whole may affect education moving forward. Many of the current systems using AI today allow for customized arrangements through the use of automated assessment tools, intelligent tutoring, and adaptive algorithms. With the use of chatbots and virtual assistants, the amount of student interaction and the speed at which students can obtain information will increase as well as hold instructors accountable for identifying learning gaps. In addition, AI tools have an enormous impact on what happens inside the classroom and in institutional management. These tools will help streamline many of the non-teaching duties or administrative assignments assigned to educators, improve how resources are distributed, and better equip school administrators and policymakers to make decisions related to education. Additionally, AI tools can be used in research practices that are unique to institutions of higher education, such as automating literature reviews, analysing trend data within a compilation of data, and contributing to predictive modelling. However, the use of AI in education has created significant ethical, technical, and sociological concerns.
Authors - Aanchal Khandalkar, Reena S. Satpute Abstract - Mobile app developments exploded lately, and it’s not hard to see why. Things like 5G, AI, machine learning, and Mobile Edge Computing aren’t just making headlines they’re actually changing how apps work. Now, apps are smarter, more personalized, and packed with features that can show up overnight. Sounds amazing, but the flip side is tough: people expect everything instantly. They want quick responses, apps that basically read their minds, and zero downtime, no matter where they are. Honestly, building apps these days is anything but simple. Hardware is a real pain for developers. Phones just don’t have the power of regular computers. You get less memory, slower processors, and, of course, batteries that bail on you before you even realize. So, developers have to work magic keep the app running smoothly without draining the battery, or else people just uninstall. And with so many apps depending on outside libraries and analytics tools, there’s a whole new pile of problems. Sure, these tools help, but they open the door to security risks. Not every team has someone who lives and breathes security, so it’s easy for privacy issues or sketchy code to sneak in. All of that puts some cracks in the process and makes building solid apps a lot trickier. Such fragmentation exists on platform that developing AI/ML application for it, on the Android platform in particular, turns out to be extremely painful to integrate on. Then it becomes a question of tradeoffs from developers’ perspective-how the usage of cloud services versus local device processing fits, on each having its share of difficulties regarding scale, speed, power and security. In this paper, we go through the workflow and delve deeper into a comparative study on native application development.
Authors - Khushi Tijare, Reena S. Satpute Abstract - Bio-signals such as eye movements, electroencephalogram (EEG), functional magnetic resonance imaging (fMRI), and pupil dilation are real-time reactions to language cues that provide more data than static text representations used in classical NLP. The aim of this study is to examine how bio signals could be employed in deep natural language processing (NLP) to enhance task effectiveness and interpretability during reading comprehension, sentiment analysis, named entity recognition (NER), and syntax parsing. The following sections will describe the important aspects of design, including. Pre-processing steps for EEG and eye-tracking datasets. Model architecture types such as feature injection, multi-task learning based on auxiliary tasks, attention mechanisms, and multimodal transformers. Experimental design and metrics used for model evaluation. Ethical considerations regarding the use of cognitive signals. All design decisions made by the authors have been verified through experiments involving open access benchmark datasets like ZuCo, ZuCo 2.0, and recently developed EEG datasets.
Authors - Luong Tran Thi, Nguyen Van Long, Bac T. Nguyen, Hiep L Thi Abstract - Maximum Distance Separable (MDS) matrices play a crucial role in the design of diffusion layers in modern symmetric cryptographic primitives such as block ciphers, hash functions, and lightweight cryptographic schemes. Owing to their ability to achieve the optimal level of diffusion as measured by the branch number criterion, MDS matrices significantly enhance resistance against differential and linear cryptanalysis. However, the practical deployment of MDS matrices often faces challenges due to high computational cost, a large number of XOR operations, and substantial hardware resource requirements. Therefore, the construction of implementation-efficient MDS matrices has become an important research direction in modern cryptographic design. This paper presents a comprehensive survey of methods for constructing and optimizing MDS matrices with a focus on reducing implementation complexity in both software and hardware environments. Specifically, classical construction methods based on Cauchy, Vandermonde, and Reed–Solomon structures are reviewed, along with structured matrices such as circulant, recursive, and involutory forms. In addition, optimization techniques targeting XOR count, circuit depth, and memory usage are analyzed, and several open research directions in the design of efficient diffusion layers for modern cryptographic systems are discussed.
Authors - Hiep. L. Thi Abstract - Recent advances in quantum computing threaten the cryptographic foundations of blockchain systems. While existing surveys analyze quantum resistant blockchain architectures at the system level, wallet-layer migration and threshold post-quantum signing mechanisms remain underexplored. This paper proposes a post-quantum secure cryptocurrency wallet architecture based on code based threshold signatures. We introduce a formal quantum-adversarial wallet model, design a modular wallet framework, provide a security reduction argu ment under syndrome decoding hardness, and outline a migration-compatible de ployment strategy
Authors - Hiep. L. Thi Abstract - Secret-sharing is a fundamental cryptographic primitive that enables the secure distribution of sensitive information among multiple participants while guaranteeing correctness and privacy. In this paper, we present a comprehensive study of secret-sharing schemes from both information-theoretic and computational perspectives. We begin by reviewing classical threshold constructions, including Shamir’s polynomial-based scheme and its algebraic interpretation via linear codes. The notions of correctness and perfect secrecy are formalized using entropy, and the role of access structures in characterizing authorized subsets is emphasized. We then examine ideal secret-sharing schemes, where each share has the same size as the secret, and highlight their deep connection with representable matroids. In particular, we discuss how matroid representability over finite fields characterizes the existence of ideal linear secret-sharing schemes, thereby linking combinatorial independence with cryptographic access control. Additional structural results concerning field dependence, minimal non-ideal access structures, and connections to linear codes are also addressed. The paper further explores computational secret-sharing, which relaxes perfect privacy to computational indistinguishability under standard cryptographic assumptions. We describe constructions based on encryption and threshold key sharing, as well as realizations derived from monotone circuits. A central theme is the separation between information-theoretic and computational models: while certain access structures require exponential share size in the information-theoretic setting, they admit polynomial-size shares under computational assumptions. Finally, we discuss algebraic methods underlying secret-sharing, including polynomial interpolation and linear coding techniques, and illustrate how these tools support efficient distributed protocols such as secure multiparty computation and threshold cryptography. Overall, the paper provides a unified treatment of structural, algebraic, and computational aspects of secret-sharing, highlighting its foundational role in modern distributed cryptographic systems.
Authors - Ioannis Patias, Koutaro Hachiya Abstract - The rapid growth of compute-intensive applications has intensified the need for selecting appropriate hardware accelerators. This paper presents a workload-to-architecture framework that explains when GPUs, FPGAs, or ASICs are most suitable, based on four determinants: parallelism granularity, memory behavior, dataflow regularity, and specialization depth. We organize representative workload classes—dense linear algebra, sparse/irregular algorithms, streaming signal processing, bit-level control workloads, and deep learning (training/inference)—and discuss how each class aligns with the execution and memory models of the three accelerator families. Our analysis highlights that GPUs excel in throughput-oriented, data-regular workloads; FPGAs provide deterministic latency via spatial pipelining and customized data paths; and ASICs achieve the best performance-per-watt for stable, high-volume tasks. The resulting framework provides practical guidance for accelerator selection and motivates heterogeneous system design.
Authors - Ronky Amber-Doh, Benjamin Ghansah, Winfred Larkotey, Stephen Opoku Oppong, Ezekiel Okoe, Olivia Osei-Tutu, Emmanuel Prah, Ephrem Kwaa-Aidoo Abstract - This paper introduces a novel framework, ExplainoGraph, that integrates square loss optimization with explainable artificial intelligence methods for knowledge graph–based recommender systems. Prior studies show that knowledge graph embeddings significantly enhance recommendation accuracy; however, they suffer from limited interpretability, thereby constraining user trust and system transparency. To address this gap, we introduce ExplainoGraph, which embeds explainability directly into the recommendation process through interpretable scoring functions and feature attribution procedures that provide meaningful insights into model decisions. Again, the framework incorporates ripple set propagation to effectively model user preferences, particularly in sparse data environments where traditional methods are suboptimal. Extensive experiments conducted on multiple benchmark datasets demonstrate that Explaino-Graph consistently outperforms the state-of-the-art baselines used across key evaluation metrics, including Precision@K, Recall@K, F1-score, and normalized discounted cumulative gain (NDCG)
Authors - Sayyora Qulmatova Abstract - This study uses machine learning models such as Multi-Linear Regression (MLR) and Holt-Winters Exponential Smoothing to model and forecast agricultural production indicators in Uzbekistan. The dataset consists of key agricultural indicators such as gross agricultural output, milk production, egg production, honey production, vegetables, fruits, and livestock products (in live weight). To improve model performance and ensure comparability, various data preprocessing methods such as StandardScaler, MinMaxScaler, RobustScaler, and Normalizer were used. The forecasting accuracy of each model was evaluated using standard error metrics such as mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE), and mean absolute percent- age error (MAPE). Empirical results show that the MLR model provides stable and interpretable forecasts, especially when combined with appropriate scaling methods. The Holt-Winters model exhibits strong performance for time series with consistent trends, but shows limitations when applied to variable data. The MLP model effectively captures nonlinear relationships and complex time patterns, although its performance is sensitive to data preprocessing, with MinMaxScaler generally yielding superior results. Overall, the results show that no model is universally optimal; instead, the choice of forecasting technique should be based on the characteristics of the data. The proposed modeling framework contributes to increasing the accuracy and reliability of agricultural forecasts and can support evidence-based policy planning and decision-making in the agricultural sector of Uzbekistan.
Authors - Salma Chlaikhy, Adil Chakhtouna, Abdellah Adib Abstract - We propose a neural encoding framework that predicts continuous ECoG signals from self-supervised Data2Vec speech representations using multivariate Ridge regression. Evaluated on 18 auditory cortical electrodes from nine participants, the model achieves a mean Pearson correlation of r ≈ 0.39 under pooled cross-validation and r = 0.29±0.018 under leave-one-subject-out (LOSO) evaluation, reaching approximately 57% of the noise ceiling. Results confirm that self-supervised speech representations capture stimulus-driven cortical dynamics, highlighting their promise for neural signal modeling and brain–computer interface research.
Authors - Raita Rollande Abstract - This paper addresses the research question: How can an effectively organized and managed adult education reskilling and upskilling process help bridge the skills gap in today’s labor market? In an era of rapid technological development and changing business needs, the demand for continuous workforce reskilling and up-skilling has become increasingly critical. This study examines a private sector-driven adult education model that prioritizes speed, quality, and flexibility in response to industry demands. TestDevLab is a fast-growing company specializing in software quality assurance. Due to the lack of industry-specific specialists from traditional higher education institutions, TestDevLab has developed an in-house training solution where experienced engineers train new specialists while also upskilling existing employees. To address these challenges systematically, the company established TDL School. This dedicated training institution has designed the Model for Rapid and High-Quality Reskilling and Up-skilling tailored to mid-sized businesses. This article presents the TDL School model, exploring innovative training approaches, curriculum development, and company collaboration to highlight best practices for effective workforce development. The findings demonstrate that a well-structured, professionally organized learning process enhances employability, strengthens industry competitive-ness, and fosters lifelong learning. By sharing this model, the study aims to pro-vide a scalable framework that other companies of similar size can adopt to ad-dress their workforce challenges. This research contributes to the ongoing discussion on adaptive education models and their role in bridging the skills gap in today’s labor market.
Authors - Isaac Baffour Senkyire, Benjamin Ghansah, Emmanuel Freeman Abstract - With the rapid development of deep learning, CNN-based medical im-age segmentation algorithms have been successful. However, study on the pancreas in 3D CT and MRI images is limited due to the excess use of computer memory and the complexity of the pancreas. In this paper, we present a memory-efficient cascaded 3D network for pancreas segmentation in CT and MRI. We develop a novel Lightweight 3D Bond (L3D-Bond) Layer to reduce filter size, and maintain performance while lowering memory usage, and a novel Light-weight 3D Asymmetric Corollary Atrous Spatial Pyramid Pooling Module (L3D-aCASPP) that captures multi-scale 3D context with lower computational cost. Our experiments were done using the public NIH pancreas segmentation dataset, MRI pancreas segmentation dataset, and MSD spleen segmentation dataset achieving a competitive segmentation performance of 80.12 DSC on the NIH dataset with parameters less than 0.5 million.
Authors - The Quan Trong, Nguyen Trong Nhan Abstract - An autocorrelation receiver can be employed in surveillance and communication systems to identify the type and operating mode of radiation sources in the absence of a priori signal information. In this work, the autocorrelation receiver is defined as a system comprising a broadband analog front-end with frequency conversion to the intermediate frequency range and a narrowband processing unit based on autocorrelation. The performance of signal processing is governed by the received pulse duration and the length of the fast Fourier transform (FFT) window. The study provides an estimate of the signal-to-noise ratio (SNR) required to achieve a specified probability of correct classification of simple radio pulses at a fixed false alarm rate. The results show that increasing the pulse duration while maintaining a fixed FFT window (i.e., reducing the ratio of FFT window length to pulse duration) decreases the required SNR. Consequently, the probability of correct classification is improved under these conditions. Furthermore, high receiver efficiency is achieved when the ratio of the FFT window length to the pulse duration is kept below 10. If the number of FFT samples is fixed, further improvement in classification performance requires increasing both the sampling frequency and the processing rate.
Authors - Uphaar Goyal, Chirag Patadia, Nitinkumar Leuva Abstract - Modern software development environments depend on cloudnative infrastructure, automated CI/CD pipelines, and distributed DevSecOps workflows. These environments improve delivery speed but expand the attack surface through privilege escalation, credential compromise, insider threats, and misconfigured pipeline permissions. Traditional Role-Based Access Control (RBAC) improves least-privilege enforcement, but static RBAC does not adequately respond to changing runtime context such as business hours, network trust, multi-factor authentication status, and deployment pipeline state. This paper proposes an enhanced cybersecurity framework that integrates RBAC, contextaware policy enforcement, and Agentic AI-based autonomous auditing. The proposed agent observes access events, learns behavioral patterns, detects anomalous requests, and generates explainable audit evidence without replacing deterministic access control. Experimental evaluation in a simulated CI/CD environment shows that the RBAC with Agentic AI framework achieves 98.21% accuracy, improves anomaly detection compared with traditional access-control models, and maintains low enforcement latency under increasing concurrency. The results indicate that Agentic AI can strengthen secure software development by adding adaptive audit intelligence while preserving the predictability and administrative clarity of RBAC.
Authors - Thokozani Nkosinathi Hlubi, Nazeer Joseph Abstract - Digital transformation is reshaping the auditing profession by introducing advanced digital tools, automation, and data‑driven processes that redefine how audits are planned, executed, and evaluated. This study examines the effect of digital transformation on auditing processes, focusing on how digital tools, automation technologies, and shifting skill requirements influence audit effectiveness and efficiency. Using a qualitative research design, Rich Picture workshops were conducted with practicing auditors to explore how emerging technologies are integrated into real-world audit environments. The findings reveal three key themes. First, digital tools enhance real‑time access to information, improve collaboration, and deepen auditors’ understanding of complex IT environments. Second, automation significantly improves audit effectiveness by streamlining routine tasks, supporting anomaly detection, and enabling more robust risk assessments—while still requiring professional judgment. Third, efficiency gains emerge through time savings, resource optimization, and evolving competency requirements, underscoring the need for continuous upskilling. Building on these insights, the study proposes a four‑stage implementation framework consisting of strategic alignment, workforce development, workflow redesign, and ethical governance safeguards. The research contributes to both theory and practice by demonstrating how digital transformation reshapes audit work and offering a structured roadmap for organizations seeking to modernize their audit functions responsibly and sustainably.
Authors - Zumna Usman, Madiha Khalid, Weiwei Jiang, Momina Shaheen, Umar Mujahid, Muhammad Najam-ul-Islam Abstract - The Internet of Things (IoT) networks operate under strict resource constraints having limited computational capability, memory, bandwidth, and energy, while still being required to combine essential security goals such as confidentiality and mutual authentication with efficiency, particularly in Radio-Frequency Identification (RFID)-based systems. For this reason, Ultra-Lightweight authentication protocols are commonly used, where traditional cryptographic techniques are often too demanding. The Random Rearrangement Block Matrix-Based Ultra- Lightweight RFID Authentication Protocol (RUAP) was introduced to strengthen security. In this paper, RUAP is analyzed and shown not to eliminate fundamental weaknesses. By applying a probabilistic disclosure attack, it is shown that public messages leak exploitable statistical information, making it possible to fully recover the identifier in reduced configurations and to recover about 71.77% of a 96-bit identifier. It is further shown that RUAP’s asymmetric key update mechanism allows adversaries to trigger desynchronization, resulting in denial of service.
Authors - Rajiv Ghai, Anil Kumar Bisht, Akash Sanghi Abstract - Agricultural supply chains face a fundamental tension consumers require transparency for provenance verification while commercial stakeholders demand confidentiality of pricing, buyer identities and quality scores. This paper introduces SelEncChain-Agri, a selective field encryption (SFE) framework that resolves these conflicting demands on a single fully decentralised Solana blockchain. Supply chain data is partitioned into public fields stored in plaintext and confidential fields encrypted with ECIES combined with Multi-Party Key Encapsulation (MP-KEM). Authorised parties decrypt fields using existing Solana wallet keypairs via a formally specified Ed25519-to-X25519 key derivation (libsodium convention), requiring no additional key material or trusted third party. Three technical contributions are made: (1) SFE field encryption with per-capsule independently randomised nonces preventing GCM keystream reuse (2) SFE decryption and (3) a key revocation protocol providing post-revocation forward secrecy. Security analysis under the Dolev-Yao model provides a constructionlevel IND-CPA argument under the DDH assumption on Curve25519. Comparative evaluation shows SelEncChain-Agri satisfies all eight stakeholder requirements from the agricultural blockchain literature, versus at most seven for dualchain alternatives. A Solana devnet prototype confirms functional correctness: end-to-end latency was 3,691 ms for batch creation and 2,111 ms for event submission; client-side SFE overhead was 74.59 ms (Python), with estimated pure cryptographic cost of 3.3 ms (Rust/WebAssembly). Keywords: Agricultural supply chain · Anchor framework · DPDP Act 2023 · ECIES · Food traceability · Key revocation · MP-KEM · Privacy-preserving · RFC 7748 · Selective field encryption · Solana blockchain
Authors - Austin jojo Jallah, Reena Satpute Abstract - As the transformer language models were developed, it was possible to introduce one of the significant technological revolutions in conversational AI. Basically, the understanding and production of human language has been enhanced by transformers. Transformer models make it possible to build dialogue systems that provide natural, complex, and context-sensitive communication to develop further customer support technologies, healthcare technologies, and virtual assistance services. In the current paper, this paper will assess advanced transformer systems, BERT, GPT, and T5, as well as their variants, assessing their performance capabilities in the use of dialogue. All the models are evaluated in terms of performance quality, which is judged by its fluency production and also gauges of coherence and contextual accuracy and its general ability to pro-cess computations. This review talks about the prompt engineering approaches and the human feedback-enhanced learning through reinforcement (RLHF) and the adapter transfer learning methods, to improve the flexibility and quality of the model. The paper presents the fresh trends in Conversational AI with multi-modal learning as well as retrieval-enhanced generation and factuality-based coherence knowledge application. Nevertheless, transformer-based models have three key weaknesses, among which, there are bias and hallucinations, and the computing requirements are very high. These weaknesses are analyzed and we discuss the potential remedies that involve symbolic deep learning combinations and efficient compression methods, such as quantification and pruning. We have found out the extent to which each system can do and the things that they cannot accomplish, and hence can help to determine the right applications of each of the frameworks. We introduce an evaluation comparison as a scholarly resource to the practitioners of dialogue system development so they can perform better with moral and effective models of operations. The article concentrates on the devel-opment of transformer-based conversational AI and its estimated impact on hu-man-computer dialogue systems
Authors - Ritesh Kumar, S. Rajaprakash Abstract - The society we live in today has seen the accumulation of knowledge via social media platforms such as Twitter and Facebook, which are expanding at a tremendous rate on a daily basis. The victims of these social media platforms are users who tweet or post on a variety of issues from any location in the globe via the usage of the internet. Tweets are used to assess both positive and negative mes-sages in order to produce polarity scores and also to have the ability to anticipate future trends. Twitter is a source from which these polarity scores may be collected; nevertheless, the information about polarity scores is kept confidential. The information will be easily compromised, and the fluctuations in the score will result in incalculable consequences, such as affecting the global economic position, the brands of corporations, and therefore the reputations of businesses. The installation of Salsa, which offers faster encryption due to the district round and greater data security, was something that Daniel Bernstein intended to do in order to address these issues. In this work, a fresh approach is provided by changing the Salsa20/4 algorithm in order to further strengthen the security of the polarity scores, which is a vital necessity in the society that we live in today. The proposed method is RRCF has two encryption stages. Stage 1 is comprised of column operations, whereas stage 2 is comprised of four procedures. Finding the greatest common factor of the pain text that has been provided is the initial step in the procedure. Identifying the time period in the pain text is the second step in the procedure. The outcome of the second step is used in the third phase, which is to create a pair of values. The application of the pair values and the swapping of the cell values in the given matrix is the fourth step. In comparison to the Salsa20/4 technique, the suggested methodology has a much higher level of security.
Authors - Boago Seropola, George Anderson Abstract - When it comes to healthcare, the implementation of machine learning (ML) and deep learning models requires a shift away from blackbox methodologies and toward frameworks that are transparent and auditable. This is necessary in order to guarantee ethical governance and patient safety. In this study, an integrated XAI-CRISP-DM framework is proposed. This methodology incorporates post-hoc Explainable Artificial Intelligence (XAI) into the iterative stages of the Cross-Industry Standard Process for Data Mining (CRISP-DM). Additionally, the research places an emphasis on continual post-deployment oversight. Within the context of HIV/AIDS risk classification in Botswana, we analyse the interpretability of non-linear decision boundaries in LightGBM and Multi- Layer Perceptron (MLP) models. This evaluation is carried out with the assistance of SHAP and LIME. For the purpose of quantitatively validating the clinical significance of socio-demographic characteristics using the publicly available dataset, the fifth Botswana AIDS Impact Survey 2021 (BAIS V), this study makes use of impact score and explanatory specificity. Results demonstrate that LIME and SHAP effectively decompose complex model outputs into human-readable feature weights, identifying key drivers such as sexual-activity-before-15 and condom use. Through the tracking of model integrity and concept drift in dynamic clinical situations, the incorporation of a monitoring stage guarantees that continuous accountability is maintained. The results of this study suggest that the XAI-CRISP-DM framework is an essential methodological standard for resource-constrained environments such as Botswana. This framework ensures that data-driven healthcare solutions are not only high-performing but also statistically reliable, equitable, and ethically sound.
Authors - Ana Laura Lezama Sanchez, Mireya Tovar Vidal Abstract - In this paper, we present the automatic classification of autoimmune skin diseases using deep convolutional neural networks. Hence in this study we conducted within the context of supervised classification of dermatological images, with the objective of designing and implementing a model capable of distinguishing among five clinical classes like lupus, psoriasis, vitiligo, lichen planus and healthy skin. Therefore, a deep convolutional neural network-based system, trained and evaluated on a labeled dataset of clinical images, is proposed. The model was evaluated using the metrics precision, recall, F1 and accuracy. The results obtained indicated that the accuracy was 70%, demonstrating the model’s ability to learn relevant discriminattive features. The best performance was observed in the healthy skin and vitiligo classes, with F1 of 0.82 and 0.80, respectively, indicating high identification capacity. On the other hand, the psoriasis and lichen planus classes showed moderate performance, with F1 values of 0.63 and 0.58, respectively. The lupus class exhibited the lowest performance, with an F1 of 0.46, reflecting the complexity of its visual variability and its similarity to other conditions.
Authors - Katherine Garcia-Velez, Daniel Maldonado Abstract - This paper examines the trajectory of digital transformation in Ecuador's public sector between 2021 and 2025. Its objective is to analyze how the country moved from an initial strategic orientation to a legal and public policy consolidation of digital transformation. Methodologically, the study adopts a qualitative approach based on documentary analysis and diachronic comparison of official instruments: the Digital Agenda 2021-2022, the Digital Transformation Agenda 2022-2025, the Organic Law for Digital and Audiovisual Trans-formation (2023), and the Digital Transformation Public Policy 2025-2030. The analysis is grounded in the idea of the progressive institutionalization of digital reform and compares the evolution of these instruments in terms of their nature, scope, and institutional implications. The findings show a four-phase sequence: agenda setting, strategic coordination, legal consolidation, and programmatic consolidation. The study concludes that Ecuador progressed from strategically oriented instruments toward a binding legal framework and a national public pol-icy. However, the existence of this institutional architecture does not, by itself, imply homogeneous results in terms of performance or service quality, so its effective implementation remains an open empirical field.
Authors - C. Bagath Basha, S. Rajaprakash, K. Karthik, Panjala Vijay Goud, Macharla Rakesh, M Ramana Kumar Abstract - The privacy and security of sensitive information is a major concern in the social assistance sector due to the widespread use of Internet of Things technologies. This study presents a secure sharing algorithm for IoT social assistance data based on blockchain and smart contracts. It aims to address the inherent hazards of conventional centralized administration, such as data loss and manipulation. This paper propose a security method and this method has four process. There are four steps to the new method. First, change the text from plain text to “ASCII code” (A). The second step is to take the A values and make pairs. Then, swap the cells in the matrix so that the even pair numbers start from the 0th cell value and go to the end of the matrix. The third step is to use the “ASCII code” as C to find the prime number. To do the fourth step, you need to use Equation 1. Take the numbers from A1 and pair them up. Then, change the cells in the matrix The message is ultimately received in its original format via the process of decryption, which is thought of as the inverse of this conversion. The proposed technique provides a higher level of security when compared to more conventional encryption methods.
Authors - Rory Lewis Abstract - This work presents a formal supervisory framework for detecting and intervening in large-scale AI misbehavior using neuromorphic sentience, supported by probabilistic guarantees. As generative and adaptive artificial intelligence systems become foundational to human decision-making, scientific discovery, and national infrastructure, ensuring their reliable and safe operation has emerged as a critical challenge. Existing approaches rely primarily on external, reactive monitoring and are insufficient for models operating at machine speed. More fundamentally, closed computational systems cannot reliably represent or act upon their own epistemic limits, creating an inherent blind spot in autonomous operation. To address this limitation, a control architecture is introduced in which an independent neuromorphic module supervises internal AI dynamics through event-driven processing and dendritic integration. Operating without a global clock, the system continuously monitors activation patterns, attention shifts, and inter-module interactions in real time, enabling low-latency and energy-efficient detection of transient and distributed signatures of instability that are inaccessible to conventional approaches. Using probabilistic inference over these signals, the framework identifies early indicators of hallucination, instability, and unintended coordination prior to output generation. Formal lemmas establish mathematical bounds on the probability of undetected misbehavior under realistic operating conditions. Finally, the framework is integrated with a human governance model in which democratically defined thresholds regulate the balance between AI capability and societal safety.
