University of the Cumberlands

This study examines how autonomous artificial intelligence agents inherit and exercise digital permissions within enterprise and cloud computing environments and evaluates the security implications of applying traditional access control frameworks to agent-driven systems. A quantitative analytical design was employed using three publicly available datasets: the Google Cloud Identity and Access Management policy dataset, the MITRE ATT&CK Enterprise dataset, and the AWS IAM Access Advisor dataset. Network centrality analysis, K-means clustering, and permission utilization ratio modeling were applied to examine permission inheritance structures, identify dominant cybersecurity risks, and assess the efficiency of RBAC and ABAC authorization models. The analysis indicates that autonomous agents possess substantially higher average permissions (63.7) than human users (11.8), while agent identities exhibit the highest network centrality (0.66). Additionally, credential propagation attacks produced the highest impact with an average data exposure of 68 GB. The findings support the need for agent-specific identity governance, dynamic permission scoping, and enhanced credential monitoring mechanisms.

The high growth rate in digital businesses has heightened issues around the world about the environmental footprint of information technology (IT) infrastructure, especially as the data centres, cloud services, and high-performance computing workloads are becoming a large contributor to increasing energy use and carbon emissions. Although the topic of sustainability has gained increased regulatory and corporate attention, there remains no set, data-based approaches to measure the ecological performance of the organizations with regard to the digital operations. The paper presents a holistic analytical model that combines sustainable IT infrastructure indicators, cloud resource optimization policy, and corporate sustainability key performance indicators (KPIs). Based on a mixed-methodology, which integrates empirical data related to industry benchmarking, cloud provider sustainability reporting, and already existing environmental reporting criteria, the framework provides a systematic approach to connect micro-level IT energy telemetry (including power usage effectiveness (PUE), server utilization rates, virtualization efficiency, and carbon intensity of workloads) to macro-level sustainability results, including reductions in greenhouse gas (GHG) emissions, integration of renewable energy, energy cost reductions, and improvement in ESG performance. Quantitative modeling based on real world data demonstrates how green data analytics can be used to aid in carbon-conscious scheduling, predict IT energy demand and optimizing allocation of cloud resources. The outcomes indicate that the incorporation of IT operating data with high-end analytics can improve greatly the level of transparency, the measurement precision, and the environmental responsibility of digital enterprises. The originality of the study is the cross-layered mapping of technical IT metrics on organizational sustainability KPIs, which can be replicated to achieve the net-zero digital transformation. The suggested framework offers practical information to businesses, policymakers, and cloud service providers interested in realizing sustainability goals in fast-changing digital realms.

This paper introduces the privacy-preserving cloud AI system, GWO-FLNet, which is an arrangement of Federated Learning (FL), Zero-Trust architecture (ZTA), and the Gray Wolf Optimizer (GWO) that allows secure and high-accuracy interaction between distributed and multi-domain systems. The first goal is to maximize the precision of the models and preserve the data privacy when the datasets of clients are statistically heterogeneous and domain-specific. GWO-FLNet utilizes a CNN-BiLSTM hybrid setup, which models spatial-temporal regularities and trains the message layer of a client without communication of raw signals between that client and the server. In order to address inter-client differences and training imbalance, GWO adjusts the learning rate, the size of BiLSTM units, and aggregation weights, personalized by the client during the federated training process. Access control and strict authentication are imposed through ZTA principles over all clients. The experimental data show that the advancement in the performance is significant as opposed to traditional FL models, including FedAvg and FedProx: average Dice Similarity Coefficient (DSC) increase of 7.8%, Intersection over Union (IoU) gain of 8.3%, and an increase in F1-score of 6.4%, besides 15% improvement in communication overhead. These results highlight the strength and scalability of the structure in situations with decentralized, nonuniform populations of data, which makes it highly appropriate for setups applicable in the real-time, privacy-conscious use of the cloud.

<title>Abstract</title> Hyperspectral imaging has become a cornerstone technology across applications such as remote sensing, precision agriculture, environmental monitoring, and intelligent surveillance, due to its ability to capture rich and discriminative spectral information. Although significant advances have been made in machine learning– and deep learning–based classification techniques, their deployment in real-world settings remains constrained. Many existing approaches exhibit limited interpretability, high computational complexity, and little to no integration of human expertise. Moreover, data-driven models often struggle in scenarios with scarce labeled samples and fail to exploit valuable domain knowledge effectively. To address these limitations, this work introduces an interpretable, human-in-the-loop framework for segmentation-based hyperspectral image classification. The proposed approach combines spectral–spatial feature fusion with affinity propagation–based segmentation and a computationally efficient Extreme Learning Machine classifier. Reliability and transparency are enhanced by embedding explainability mechanisms and structured expert feedback directly within the learning process. In contrast to fully automated pipelines, the framework allows human experts to assess uncertain predictions and iteratively refine the model through guided feedback. Rigorous mathematical formulations are presented to describe feature integration, similarity computation, classifier training, and feedback-driven optimization. Extensive experiments on widely used hyperspectral benchmark datasets demonstrate that the proposed framework consistently outperforms conventional classification methods, particularly under limited training data conditions. Performance gains are evident not only in classification accuracy but also in robustness and interpretability. These results highlight the effectiveness of integrating segmentation techniques, lightweight learning models, and human-centered AI principles to build reliable hyperspectral classification systems. Overall, the proposed solution provides a scalable, transparent, and practical approach for real-world applications where expert oversight and explainability are critical.

This study examines the convergence of rapid Central Bank Digital Currency adoption and the growing threat that quantum computing poses to existing cryptographic systems. As digital sovereign currencies expand globally, with many economies actively exploring or piloting CBDCs, the vulnerability of classical public key cryptography to quantum attacks creates an urgent need for quantum resistant solutions. The research responds to this challenge by developing an evaluation framework focused on secure and interoperable cross border CBDC systems in a post quantum environment. Using desk based computational analysis and simulation methods suitable for remote researchers, the study evaluates lattice based algorithms standardized by the National Institute of Standards and Technology, specifically ML KEM for key encapsulation and ML DSA for digital signatures. The analysis confirms strong quantum resistance against known quantum algorithms, effectively removing harvest now decrypt later risks for future transactions. Interoperability assessments show that ISO 20022 with cryptographic abstraction is well suited for retail CBDC systems, while bridge based architectures such as BIS mBridge are more effective for wholesale settlements. Performance testing on thousands of simulated transactions indicates that post quantum protocols are feasible for wholesale use but introduce significant computational and bandwidth overheads that limit immediate retail scalability without hardware acceleration. Overall, the findings support a phased migration strategy that combines classical and post quantum cryptography in the near term, guided by cryptographic agility. The study provides practical insights for central banks seeking to balance security, efficiency, and cross border compatibility in an evolving quantum threat landscape.

Aim: This paper examines the potential to improve the early-stage performance of pharmaceutical product launches, particularly within the critical first 3 - 6 months, through the use of supervised machine-learning classification models. Methods: The study employs supervised machine-learning models, including logistic regression, random forests, and neural networks, implemented on the SAP Business Technology Platform (BTP). These models integrate SAP S/4HANA transactional data with external market indicators to optimize launch strategies by improving prescriber targeting, detecting market-access risks, segmenting patient and payer profiles, and enabling real-time decision-making. Results: The findings indicate that the models decrease forecast variance, improve prescriber reach, and enable earlier corrective interventions during the early launch stage. The models support managerial decision-making by scoring prescriber adoption propensity and categorizing market-access risks to streamline payer strategies. Conclusion: The study demonstrates that integrating predictive classification models into ERP-based pharmaceutical workflows enhances agility and supports more evidence-based, accurate, and timely decision-making, thereby improving overall launch performance. Recommendation: The study highlights the need to integrate supervised machine-learning models into pharmaceutical launch workflows while emphasizing the importance of governance systems, privacy, equity, and regulatory considerations to enable proactive changes, improve market accessibility, and optimize resource allocation.

The growing number of global shocks, which include geopolitical instabilities, pandemics, and climate shock, among others, has revealed major weaknesses in the U.S. manufacturing supply chains. The need to increase supply chain resiliency and, at the same time, improve sustainability and resource efficiency has become a strategic necessity. The proposed study is an AI- and machine learning-controlled predictive analytics-based framework created to enhance the reliability and resilience of the U.S. manufacturing supply chain systems. The framework incorporates heterogeneous data sources such as operational, environmental, market, and logistics data to facilitate the proactive identification of risks, prediction of demand, and optimization of resources. Developed machine learning models are used to embrace non-linear relationships with complexities across the supply chain nodes to assist in real-time decision-making and adaptive response strategies. The proposed framework enables the identification of disruption early in the supply chain planning and execution, better management of inventory and capacity, reduced waste, and increased energy and material efficiency by integrating predictive analytics into the supply chain planning and execution. In theory, this study will be a step forward in the interplay of artificial intelligence, sustainable supply chain management, and resilience engineering. In practice, it provides decision-makers with a scalable and evidence-based instrument in order to enhance the continuity of operations, environmental performance, and competitiveness in the long run of the U.S. manufacturing supply chains.

Strong defence mechanisms are necessary to defend the availability and integrity of network infrastructure in the ever-changing cybersecurity landscape, where Distributed Denial of Service (DDoS) attacks are becoming more common. The capacity of Deep Learning (DL) models to automatically learn feature representations and discern complicated patterns within network traffic data has made them a potential strategy for DDoS attack finding and mitigation. The design of a critical cybersecurity threat that disrupts network operations and causes considerable economic losses internationally also affects how well DL models fight against developing attacks. The Shepard Interpolation Neural Network (SINN) classifier, fine-tuned with the Artificial Rabbit Optimiser (ARO), is proposed as a new method for DDoS assault detection in this paper. The technique takes advantage of the Kaggle DDoS dataset, which is naturally skewed and poses problems for conventional ML models. The solution is to use the Synthetic Minority Oversampling Technique (SMOTE) to level the playing field in the dataset, which will result in better model presentation in every class. Thanks to ARO’s hyperparameter optimisation, the SINN classifier—which is already well-known for its interpolation capabilities and resilience when dealing with nonlinear patterns—achieves even better detection accuracy. Experiments show that the suggested model outdoes state-of-the-art machine learning tactics for DDoS attack finding in terms of accuracy, recall, and F1-score. Reliable classification results are contributed to by the incorporation of SMOTE, which successfully mitigates bias induced by class imbalance. In addition to establishing a scalable framework for practical applications, this work showcases SINN-ARO’s capabilities in protecting networks from ever-changing cyber threats. Investigating how well this approach works in real-time, dynamic network settings and expanding its applicability to identify different forms of cyberattacks are the primary goals of future studies.