In this research work, multiple machine learning regression techniques were used to predict the pollution and offer a comparative study to establish the optimum model for reliably predicting air quality in terms of data quantity and processing time. The Mean Absolute Error (MAE), Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) were used as evaluation measures to compare these regression models. Furthermore, the processing time of each algorithm was determined via standalone learning and hyper-parameter tweaking to produce the best-fit model in terms of computational time and error rate. In this paper, we have calculated the custom score which is sum of MAE, RMSE, MAPE and time processing values. The best model obtained is the custom stacked regression model has custom score of 111.41 which is very less as compared to other regression models.
The preservation of rare documents in the form of image collections presents significant challenges regarding access to their documentary content. To enable this accessibility for software agents, this article proposes a formal representation of this type of document through a semantic description layer. This layer includes a set of descriptive metadata attached to the document, alongside the minimal and strictly necessary vocabulary required to formalize the explicit textual and visual knowledge of its documentary content. To achieve this, we present a construction methodology based on a Semantic Model of Document (SMD), where a document is treated as a core documentary resource containing a set of information resources. The semantic description of these resources, aligned with RDF framework logic, produces an Ontological Core of Document (OCD) that formally describes the document's logical structure and captures its underlying semantics. Finally, we demonstrate the practical utility of these Ontological Cores through three distinct use cases—each targeting a specific dataset level (structural, administrative, and semantic)—showing how they allow software applications to move beyond simple collection searching toward intelligent, precise information extraction directly from the documentary content.
The COVID-19 pandemic catalyzed an unprecedented surge of misinformation on social media, frequently intertwined with emotionally charged language. Understanding both the sentiment and truthfulness of this content is critical for public health monitoring and misinformation mitigation. However, Bangla—despite being a globally prominent language—remains severely underrepresented in joint sentiment and fake news detection research, with existing studies largely restricted to single-task settings. To bridge this gap, this paper proposes a novel multi-task BanglaBERT-based framework for the simultaneous classification of sentiment and truthfulness in COVID-19 discourse. Furthermore, we introduce the first publicly available, dual-annotated Bangla corpus for this domain, comprising 35,526 textual samples aggregated from social media and news sources. Our architecture employs a shared BanglaBERT encoder with dual task-specific heads, optimized using a task-prioritized loss function that combines modified Focal Loss and weighted cross-entropy to address inherent class imbalances. Extensive experiments demonstrate that the proposed model achieves 75.1% accuracy (Macro F1: 0.707) for sentiment classification and 88.0% accuracy (Macro F1: 0.851) for truthfulness detection. Ablation studies and error analyses confirm that our tailored loss strategies significantly enhance the recognition of underrepresented and semantically ambiguous classes, particularly neutral sentiments. By releasing our dataset, code, trained models, and a Gradio-based interactive demo, this work establishes a robust benchmark for multi-task learning in low-resource Bangla NLP and provides a practical tool for fact-checking during health crises.
This study investigates the impact of AI-powered digital assistants on students’ feelings, engagement, and academic success within higher education environments. The study aims to investigate post-adoption behaviour, emphasizing how service experiences, functional attributes, information quality, and ease of interaction influence emotional and behavioural results. A structured survey was used to gather data from 431 respondents in higher education at Exploits university, Malawi, and the study utilized a quantitative approach. Measurement scales were adapted from validated studies in the AI adoption and educational technology literature and contextualized for the higher education setting. Partial Least Squares Structural Equation Modelling (PLS-SEM) version 4.1.1.8 was utilized to examine the connections between variables. Common method bias was assessed using the full collinearity approach, and all VIF values were below the recommended threshold, indicating that common method bias was not a significant concern. The results indicate that Service experience leads to Positive emotions (β = 0.205, p = 0.002) and Student engagement (β = 0.242, p < 0.001), validating H1a and H1b. Quality of information → Positive feelings (β = 0.161, p = 0.005), backing H3a, whereas Functional characteristics → Student involvement (β = 0.409, p < 0.001), supporting H4b. Positive emotions → Student involvement (β = 0.190, p < 0.001) and Academic achievement (β = 0.460, p < 0.001), and Student involvement → Academic achievement (β = 0.310, p < 0.001), confirming H5–H7. Contextualization × Positive emotions → Academic performance was noteworthy (β = 0.063, p = 0.038), reinforcing H8a. Nonetheless, H2a, H2b, H3b, H4a, and H8b received no support (p > 0.05). The research advances theoretical understanding by broadening AI adoption literature to include emotional and behavioural effects, while also enhancing practical implications by highlighting service quality and system efficiency. Suggestions emphasize the importance of focusing on contextual, high-quality AI resources to enhance student engagement, emotional well-being, and educational achievement. The results demonstrate that service experience is the most influential antecedent of both emotional and behavioural outcomes, whereas the effects of functional features and information quality vary across the examined relationships.
There’s a lot of promise around artificial intelligence for education to personalize learning; however, there has been very little research regarding Artificial Intelligence (AI) applications in fields with very few resources for its implementation. This paper describes a proposed AI-based adaptive learning system that aims to personalize STEM education in a low-resource school environment in Kenya. This research addresses the numerous challenges associated with such a system, such as irregular internet access, limited computer hardware in situ and no previous teacher background in both AI and education. To address these issues, an adapted reinforcement learning algorithm will personalize the content shown to students, and a modified liquid neural network is used for the prediction of student success, while not being computationally expensive. As compared to traditional adaptive systems, this adaptive learning platform supports edge computing and offline updates in order to operate in a consistently low connectivity environment. In addition, by continually adjusting the difficulty, format, and rate of delivery of STEM topics to fit the style, prior knowledge, and attention level of the individual student, this platform has been seen to improve the educational experience. An 8-month case study was done with 6 Kenyan low-resource schools and 6 comparison schools located in the city. In this case, we report a 31% increase in the level of understanding of students’ key STEM subjects, 27% reduction in student drop rate from STEM topics, and 43% increase in teacher efficiency over traditional methods. We were also able to predict the level of performance of students to a 89% success while occupying a low 1.9MB memory, making it feasible to be employed on budget Android devices. This study presents evidence that the utilization of AI for personalized adaptive learning technologies in order to minimize the disparities in the provision of STEM education in low-resource settings worldwide is possible.
Learning management systems (LMS) have become an important part of the modern higher education industry, and Moodle is one of the most popular open-source systems, which have become popular on an international level. Despite its extensive application, the majority of the existing evaluation tools are inclined to examine either performance, sustainability, or scalability separately and, thus, cannot be useful in long-term institutional planning. The current study proposes an assessment model that is a rational model of a Moodle-based LMS, as all three dimensions are included in the assessment model. The framework identifies the performance of the system, resource utilization and resource management, load-adaptive scalability metric, and an adaptable and predictable algorithmic-based assessment procedure in a variety of deployment environments. The framework applies Min–Max normalization and weighted aggregation to combine the three evaluation dimensions into a unified assessment score. To make the model more realistic, it was tested on publicly available data, for example, the Open University Learning Analytics Dataset (OULAD), which simulated actual interactions between LMS users and the utilization of cloud resource traces to analyze scalability and sustainability. Experimental evaluation using the OULAD and cloud resource datasets demonstrated approximately a 10% improvement in performance under medium workload conditions, an 18% reduction in sustainability due to increased resource utilization, and a 20% improvement in scalability as workload increased. These findings demonstrate that the proposed framework provides a systematic and data-driven approach for evaluating Moodle-based learning management systems and supports informed institutional decision-making.
The proliferation of online misinformation demands the development of highly accurate and computationally efficient automated systems for Fake News Detection. A primary impediment to system performance is the high dimensionality of textual features derived from techniques like TF-IDF, making optimal Feature Selection a critical step. This paper presents a detailed comparative experimental study of two prominent bio-inspired evolutionary metaheuristics, the Genetic Algorithm (GA) and Particle Swarm Optimisation (PSO) used as wrapper-based FS techniques for FND. The methodologies were rigorously tested across two distinct textual datasets: the complex, large-scale FakeNewsNet corpus and a moderate-scale general news dataset. The feature sets, once optimised, were evaluated using six standard Machine Learning (ML) classifiers. The GA-based FS approach, emphasising global exploration, achieved state-of-the-art accuracy of 99.91% with the Random Forest classifier on the FakeNewsNet dataset. In contrast, the PSO-based FS approach, valued for its rapid convergence, yielded a maximum accuracy of 93.29% with the Support Vector Machine (SVM) on the general news dataset. This analysis provides empirical evidence of the intrinsic trade-off between the algorithms: GA is superior for maximising accuracy in high-dimensional, complex textual spaces, while PSO offers a more efficient and practical solution for resource-constrained or moderate-scale FND tasks. The study confirms that evolutionary computation provides a robust, effective pathway for significantly enhancing ML classifier performance in this critical domain.
The aim is to design a comprehensive shelf space optimization framework that maximizes profitability, enhances sales forecasting, improves efficiency of inventory management, and supports effective decision-making in retail businesses. A robust and interactive analytical dashboard is developed that allows users to visualize critical sales metrics, analyze historical data trends, and accurately forecast product demand and supply requirements based on seasonal variations and sales performance. The work integrates three mathematical optimization paradigms like Linear Programming (LP), metaheuristic search via Genetic Algorithms (GA), and reinforcement learning using Proximal Policy Optimization (PPO) to support both static and adaptive allocation strategies. Experimental validation highlights the relative advantages of each method, with detailed evaluations based on forecast accuracy, inventory turnover efficiency, shelf utilization rate, and overall improvement in profitability. Unlike traditional static optimization models, the PPO-based framework continuously adapts allocation decisions using environmental feedback, improving flexibility in dynamic retail scenarios The paper uses multi-objective shelf optimization considering profitability, utilization, and customer demand simultaneously. The results demonstrate that the integration of predictive analytics and advanced optimization techniques significantly performs traditional shelf management approaches, offering retailers actionable insights and operational advantages.
The proliferation of massive IoT networks has created an environment where distributed AI can be achieved. At the same time, it introduces serious privacy and security challenges. Federated learning (FL) allows training local models on IoT devices and aggregating them without sharing data, but still suffers from problems such as gradient inference attack, Byzantine model poisoning attack and the failure in single point of failure centralized aggregation point. In this paper, we propose QFL-BC, a framework combining Quantum Key Distribution (QKD) and a permissioned blockchain to holistically tackle the problem. Using the BB84 protocol with decoy states, QKD generates a One-Time Pad key to encrypt the model update and achieve information-theoretic security with provable security against a quantum attacker. The central aggregator is replaced by the permissioned blockchain with a smart contract, which ensures an immutable audit trail and distributes the orchestration of FL training decent rally, as well as imposes a penalty on malicious participants by automatic reputation score maintenance. The experiments with MNIST and CIFAR-10 on 100 IoT clients under Non-IID conditions show QFL-BC obtains an accuracy of 96.8% against 41.5% for classic FL under 10% poisoning attack (133% relative improvement). We have tested its robustness across adversary percentages of 10%-40% with accuracy above 87.3% and measured scalability up to 500 clients, showing good degradation, communications overhead of 5.84 MB per round, which is only 12.3% higher than the classic FL and analysed latency and energy to evaluate its feasibility on resource-constrained IoT devices.
Sungai Kunjang is one of the primary land transportation facilities in Samarinda City, East Kalimantan, located on Untung Suropati Street in the Karang Asam Ulu subdistrict. Officially inaugurated on June 24, 1989, by Mayor Waris Husain, it serves multiple transportation modes, including public passenger vehicles (PPV), pioneer services, and intercity routes. Although the station currently provides essential information services—such as departure schedules, route options, fare details, and a basic complaint system—these services are not yet supported by a structured Information Technology (IT) and Information System (IS) framework. The lack of integration hampers service efficiency and the optimization of business processes. This research aims to design an Enterprise Architecture (EA) for the station by applying The Open Group Architecture Framework (TOGAF) Architecture Development Method (ADM). The proposed design focuses on aligning business objectives with IT/IS strategies to improve the delivery of transport information and complaint management services. The resulting blueprint is expected to serve as a strategic reference for developing an integrated information system that enhances decision-making, streamlines operations, and improves service quality at Sungai Kunjang Station. By using selected phases of the TOGAF ADM, the study provides a practical foundation for digital transformation within public transport infrastructure in the region.
This research implements an intelligent, bilingual pregnancy health monitoring system for expectant mothers. A significant problem commonly experienced by expectant mothers in rural areas in Nigeria is the unavailability of a decent antenatal system and a shortage of experienced medical personnel and equipment. The proposed system comprises IoT sensors, including Electrocardiogram (ECG), body temperature, and heart rate sensors, connected to an ESP32 microcontroller for data acquisition and transmission. A predictive system built using Random Forest and Support Vector Machine (SVM) classifiers categorises pregnancy risk into low, medium, and high. A Flask-based web application for real-time data visualization and diagnosis was developed to display the collected data and visually represent the risk level diagnosis. The performances of the predictive models, Random Forest and Support Vector Machine (SVM), were evaluated using accuracy, precision, recall, and F1-score. Random Forest achieved an accuracy surpassing SVMs by a margin of 5.28%. Random Forest and SVM precision were then compared and there was an improvement of 6.49%. In addition, Random Forest had a higher recall than SVM by 6.58%, and also had a performance increase of 6.49% on F1-score as compared to SVM. The comparative analysis shows that the Random Forest model works better than SVM in all the main measures. In this project, the Random Forest model was better than the SVM because it uses ensemble learning to manage the non-linear relationship, imbalance data and noise better to achieve superior accuracy, recall, and the F1 Scores. It was also more reliable in categorizing risks in pregnancy, as it was interpretable, which was also strong and guaranteed the timely and suitable intervention of health care
This research aims to develop and assess an information system for managing daily operations in the palm oil plantation industry, specifically for PT Kaltim Utama Plantation I. As a crucial part of Indonesia's economy, the palm oil sector faces numerous operational challenges, such as tracking activities, monitoring resources, and generating accurate reports. The designed system focuses on enhancing processes like spraying, land clearing, fertilizing, and harvesting. The study utilizes the System Development Life Cycle (SDLC) methodology, following the waterfall model, which includes the stages of analysis, design, implementation, testing, and maintenance. In the analysis phase, system requirements were identified to meet operational demands. The design phase involved creating workflows and data structures to represent plantation operations. The outcome is a web-based information system prototype that streamlines management processes, centralizes data, and improves reporting precision. The system is aimed at simplifying information retrieval, increasing operational efficiency, and aiding decision-making. The implementation is expected to address operational inefficiencies at PT Kaltim Utama Plantation I by reducing manual processes. Additionally, the system includes data visualization tools to quickly evaluate plantation performance and offers scalable features for future growth.
Sustainable grazing management requires balancing livestock productivity with ecosystem preservation, yet existing monitoring systems integrate heterogeneous data from IoT sensors, satellite imagery, and field surveys without a unified semantic layer, limiting holistic decision support. This paper proposes ONTOGRAZING, an ontology-based monitoring architecture for sustainable grazing management. Using the Uschold and King ontology engineering framework, domain knowledge was collected through surveys involving 23 livestock farmers and 4 agro-pastoral institutions in Cameroon, complemented by a systematic literature review. Seven core concepts and fourteen semantic relationships were modeled in OWL using Protégé. A five-module monitoring architecture composed of Query Reformulator, Data Integrator, Source Monitoring, Alert, and Storage modules was designed around the ontology. ONTOGRAZING was evaluated using the HermiT 1.4.3.456 reasoner and SPARQL queries. The ontology contains 47 classes, 14 object properties, and 9 data properties, and passed all consistency checks. Comparative analysis demonstrates that ONTOGRAZING is the first ontology to jointly cover forage management, dietary preferences, pasture composition, ecological–economic trade-offs, and land-use regulations. These results highlight the potential of ontology-based integration to improve interoperability and semantic decision support in agro-pastoral systems, while future work will focus on full prototype implementation and integration with real-world IoT platforms and agricultural databa
Edge computing has emerged as a critical paradigm for enabling low-latency, bandwidth-efficient, and scalable data processing in distributed IoT environments. However, its effectiveness fundamentally depends on how data is cached, stored, aggregated, and fused across heterogeneous and resource-constrained edge nodes. To address this, the present survey conducts a comprehensive and methodologically rigorous examination of data-management techniques in edge computing. An initial corpus of 150 publications was collected from major scientific databases and processed through the PRISMA framework, resulting in 25 high-quality surveys that revealed data management as the most fragmented and underdeveloped component of the edge ecosystem. Building on these insights, we performed an in-depth analysis of 75 state-of-the-art research papers published between 2018 and 2025, covering four core data-management pillars: data caching, data storage, data aggregation, data validation and data fusion. For each area, we synthesize current design strategies, highlight measurable performance outcomes, and critically evaluate architectural, algorithmic, and system-level limitations. A unified cross-technique analysis further reveals unresolved challenges in scalable data placement, coded storage, privacy-preserving aggregation, multi-modal fusion, and the absence of integrated data pipelines. The survey concludes by outlining open research directions and proposing a consolidated roadmap toward intelligent, interoperable, and workload-aware data-management frameworks for next-generation edge computing systems.
Information and Communication Technology (ICT) has become central to teaching and learning in higher education, yet effective integration depends on users’ ICT competence and institutional implementation strategies. This study examines the role of ICT literacy in supporting teaching and learning and evaluates whether cultural perceptions significantly influence ICT adoption among undergraduate students at Kaduna State University, Nigeria. A quantitative survey design was employed, with data collected from 150 students and academic staff and analysed using descriptive statistics and the Mann–Whitney U test. The findings suggest that ICT literacy is perceived as positively supporting access to learning resources, communication, and classroom engagement. However, statistical tests did not reveal significant differences between respondent groups, indicating broadly similar perceptions among students and staff. While respondents acknowledged cultural considerations, these factors did not exert a substantial influence on ICT acceptance or use within the institutional context. The results suggest that ICT literacy and institutional support structures play a more immediate role in shaping technology adoption than general cultural perceptions. From a management engineering perspective, the study highlights the importance of ICT infrastructure planning, structured digital skills training, and policy-driven ICT integration in higher education. The findings provide practical guidance for university administrators and education managers seeking to improve the effectiveness and sustainability of ICT-enabled teaching and learning systems.
Identity and Access Management (IAM) is critical for securing digital assets, particularly in financial technology (FinTech) systems, where unauthorized access can lead to significant financial losses. Three formal research questions guide this work: (RQ1) Do AI-driven models statistically significantly outperform traditional rule-based IAM systems in anomaly detection accuracy? (RQ2) Which AI model best balances precision and recall for real-time insider-threat detection under class-imbalanced IAM log conditions? (RQ3) Are the observed performance gains robust and stable across cross-validated experimental folds? This study evaluates the performance of AI-driven anomaly detection models, including autoencoders, random forests, and support vector machines, in detecting unusual user activities and potential insider threats. The Autoencoder model achieved the highest overall accuracy of 94.2% (+/- 0.8% across five-fold cross-validation) with a precision of 92.8% and recall of 91.5%. The Random Forest attained a slightly lower accuracy (92.5%) but excelled in recall (93.2%), highlighting its strength in identifying actual malicious activities. Compared to traditional rule-based IAM methods, which achieved only 78.4% accuracy, AI models significantly improved anomaly detection, particularly for subtle or previously unseen threats. McNemar's tests confirm that all accuracy improvements over the baseline are statistically significant (p < 0.001). The Autoencoder also demonstrated the lowest latency (120 ms), making it suitable for real-time deployment. These results confirm that AI-enhanced IAM systems can effectively strengthen security and operational efficiency in FinTech environments, within the scope of the simulated and publicly available datasets employed in this study.
This study presents RiceVision, a cross-platform software system for real-time rice variety identification using deep learning–based image analysis. Unlike prior work that primarily focuses on classification accuracy, RiceVision emphasizes reproducibility, deployment, and usability in real-world agricultural environments. The system integrates a web-based platform and an offline-capable Android application within a unified architecture, ensuring consistent preprocessing and inference across platforms. Deep learning models are deployed using TensorFlow and TensorFlow Lite to support both online and on-device inference. The proposed hybrid framework combines convolutional neural networks (CNNs) and Vision Transformer (ViT) architectures using a stacked ensemble strategy. Experimental evaluation on a 62-class rice variety dataset demonstrated strong classification performance, where the stacked ensemble achieved an average 5-fold validation accuracy of 98.64%, outperforming individual VGG16 (90.64%) and ViT-B/16 (91.28%) models. The system further demonstrated stable convergence behavior and low inter-fold variance, indicating robust generalization capability. A centralized model management mechanism enables version control and seamless updates across deployment platforms. Detailed model configurations, validation results, and explainability analyses are provided in the Supplementary Material. RiceVision highlights the potential of deployable AI systems for practical decision support in digital agriculture.
This paper introduces a novel approach an AI-powered Multi-Agent System (MAS) for dynamically optimizing support to enhance real-time travel reservation-side customer experience. It has an architecture with specialized agents working together under a centralized agent manager, including natural language processing, booking, optimization, and context-aware modules. The system proposes to address common constraints encountered in traditional travel platforms: delayed response to user queries, ambiguity treated poorly, and adaptation to user preferences not incorporated. Through simulated environments and realistic use cases, the MAS enables complex travel requests to be dealt with, availability to be changed dynamically, and user satisfaction to be enhanced. The modular architecture design allows easy integration into larger smart tourism infrastructures. This study thus pushes the frontier further by merging AI, multi-agent collaboration, and user-centered design in a time-sensitive application world. Future directions include adaptive learning agents, multilingual interaction capabilities, and broadening the domain applications to hotel management and intelligent itinerary planning.
Alzheimer disease is a chronic neurodegenerative disorder and the primary cause of dementia among the population, which has a huge burden to the patients, their caregivers and the health care system. Timely intervention is necessary to reduce disease progression, facilitate timely intervention and improve the quality of life. But the traditional forms of diagnostic are frequently costly and non-available especially in resource-deficient environments. The research paper proposes an interpretable and cost-efficient machine-learning model that can be used to identify the presence of Alzheimer disease at its early stages based on clinical and demographic metrics based on the Open Access Series of Imaging Studies cross-sectional dataset, which contains 436 participants. The data consists of seven numeric and two categorical variables, whereas the Clinical Dementia Rating was changed into two categories namely demented and non-demented. An extensive preprocessing pipeline was used, which entailed missing value imputation, categorical encoding and elimination of irrelevant variables, as well as class balancing with the Synthetic Minority Oversampling Technique. A number of machine learning models were tested, which comprise Logistic Regression, Support Vector Machine, Random Forest, Gradient Boosting, and Extreme Gradient Boosting. The results show that the highest accuracy of 92% was attained using the model implemented by the ensemble and the tree, with the most accuracy being returned by the Random Forest and the ensemble model. Random Forest, too, had a sensitivity of 95%, whereas Gradient Boosting and Extreme Gradient Boosting had the highest area under the receiver operating characteristic curve of 98%. The models were implemented as a lightweight web application on the Flask framework, which can make real-time predictions and color coded. The system illustrates the possibility of combining interpretable machine learning with web technologies to make it possible to conduct easy and effective early screening of Alzheimer disease under resource-limited healthcare conditions.
One of the effects of the rapid adoption of the cashless policy in Nigeria and the introduction of new naira notes is operational difficulties among financial institutions, which have led to a significant increase in ATM card theft and fraud among clients. Absence of real-time analysis of access points, combined with the intermittent and simultaneous quality of fraudulent dealings, are two major factors that make conventional fraud detection systems fail regularly. Towards reducing ATM fraud, this paper will present a high-performance, intelligent based, AI-based model to integrate three factors of biometric authentication, spending pattern analysis, and password verification into a three-factor model. Results of experiments based on real banking data prove that the proposed solution is superior to traditional models in terms of accuracy, precision, recall, and F1-score. The model uses an optimized Bi -Directional Long Short-Term Memory (BiLSTM) network to analyze historical ATM transaction records and identify behavioral abnormalities that could point to fraud. A Cuttlefish Optimization (MCFA) algorithm that is based on mapping is used to fine-tune the parameters, thus improving the reliability and accuracy of the classification. Biometric verification combined with behavioral modeling using AI stands out as a scalable and dependable framework of minimizing ATM card fraud and instilling confidence within the banking industry
