ResearchBerg
Not a member yet
161 research outputs found
Sort by
Graph Neural Network for Service Recommender System in Digital Service Marketplace
Get PDFThe emergence of the platform economy has resulted in the decline of many traditional forms of doing business. Freelance work makes use of a platform to connect businesses or people with other businesses or persons in order to solve particular issues or deliver specific services in return for payment. The pairing process involves a buyer that needs work done, a platform that handles the algorithm, and a worker who is willing to do the job via the platform. This research argues that by efficiently pairing the talents of workers to the requirements of buyers, the platforms have the ability to expedite business operations for buyers, empower platform workers, and significantly improve the overall customer experience. Graph Convolutional Networks (GCNs) are inspired by CNNs and aim to expand the convolution operation from grid records to graph records, which in turn facilitates advances in the graph domain. In order to develop reliable and accurate embeddings for digital service recommendation, we employed a graph-based technique on a freelance platform dataset using the graph linkages of services and buyer data. We employed an aggregation-based inductive graph convolution network, namely, Graph SAmple and aggreGatE (GraphSAGE). It is a generalized inductive architecture that learns to construct embeddings for previously unknown data by sampling and combining attributes from a node\u27s immediate neighborhood. We also applied PinSage, a stochastic Graph Convolutional Network (GCN) that can learn node embeddings in platform networks with many digital services. When a robust recommender system is used in digital service marketplace, it can offer promising results that may increase users\u27 satisfaction with the service and boost the platform\u27s ability to increase revenue
UNCOVERING EVIDENCE OF ATTACKER BEHAVIOR ON THE NETWORK
Get PDFThis comprehensive research presents and investigates a diverse assessment of interruption discovery strategies and their job in contemporary online protection. Interruption Recognition Frameworks are taken apart as vital parts in defending computerized foundations, utilizing different techniques, for example, signature-based, peculiarity based, and heuristic-based identification. While signature-based strategies demonstrate strong against known dangers, the review highlights the urgent job of irregularity-based and heuristic-based approaches in countering novel and complex assaults. Different types attract, their characteristics and behaviors has explored in this paper. The mix of AI and Man-made consciousness (computer based intelligence) in recognizing odd exercises arises as an extraordinary power, empowering versatile reactions to developing digital dangers. The exploration fundamentally breaks down the difficulties looked by existing location strategies, including versatility concerns, high bogus positive rates, and the encryption-related obstacles in rush hour gridlock examination. The outcomes and investigation segment approves the viability of proposed models, including group learning strategies and creative techniques, for example, the Solid Methodology in light of Blockchain and Peculiarity based location (SABA). A Convolutional Brain Organization (CNN) model for interruption location in IoT conditions and a cross breed approach joining positioning based channel strategies and NSGA-II exhibit eminent exactnesses. The review\u27s suggestions for network security are significant, prompting proposals for a TTP-driven approach, mix of conduct peculiarities, persistent security mindfulness preparing, standard red group works out, versatile episode reaction plans, and intermittent security reviews. By and large, the examination contributes a nuanced comprehension of assailant\u27s ways of behaving, down to earth procedures for online protection flexibility, and makes way for future investigation into dynamic danger scenes and the human component in network safety
An Adaptive Hybrid Ensemble Intrusion Detection System (AHE-IDS) Using LSTM and Isolation Forest
Get PDFNetwork intrusion detection is vital for the security of today\u27s computer networks against malicious behavior. Existing detection systems often fail at achieving a good trade-off between the detection of known patterns of attacks and the detection of novel, unseen attacks. In this paper, we propose an Adaptive Hybrid Ensemble Intrusion Detection System (AHE-IDS) based on a supervised deep learning detector and an unsupervised anomaly detector to enhance detection rates. The suggested system integrates an Isolation Forest outlier detection technique and a Long Short-Term Memory (LSTM) neural network in a complementary way. The LSTM is trained with normal and malicious traffic patterns over the CSE-CIC-IDS2018 benchmark dataset that has a wide variety of attack scenarios. While the LSTM is perfroming this, the Isolation Forest is learning normal patterns to detect anomalies that could be new intrusions. A weighted voting method that adaptively weighs the two models\u27 outputs dynamically combines the outputs of both models in such a way that the ensemble is able to prioritize the more dependable detector in accordance with prevailing conditions. Both known attacks and previously unseen anomalies outside standard traffic patterns can be detected using this hybrid method. AHE-IDS is tested on the CSE-CIC-IDS2018 dataset. Experiment results demonstrate that the ensemble achieves low false alarm rates and high detection rates, outperforming both LSTM and Isolation Forest individual models. According to the findings, AHE-IDS greatly enhances the accuracy and recall of an individual LSTM classifier at a low false positive rate. It successfully decreases missed attacks without influencing precision. The adaptive weighting scheme improves robustness as it adapts to concept drift and changing attack patterns over time. Consequently, AHE-IDS performs well in dynamic environments. The system is adaptive that can react to both familiar and unfamiliar kinds of cyber attacks
Dynamic Cybersecurity Strategies for AI-Enhanced eCommerce: A Federated Learning Approach to Data Privacy
Get PDFAI in eCommerce has implemented machine learning, natural language processing, and more recently advanced to optimize recommendations, pricing, and content for better personalization of customer experiences. The more personalized the user\u27s experience, the greater their exposure to various cybersecurity threats around data breaches, adversarial manipulations, and unauthorized account access. This paper explores adaptive cybersecurity protocols that can protect these AI-driven personalization systems without sacrificing their overall effectiveness. We also touch on context-aware authentication and access control, such as risk-based adaptive authentication and zero-trust architecture, which add to security measures responsive to users\u27 real-time behavior. It examines methods that will protect data and AI models from leakage and model exploitation using federated learning, homomorphic encryption, and differential privacy. We analyze AI-powered anomaly detection techniques that help in rapid identification and response to threats, and secure API management practices that prevent interface abuse for communication. Great emphasis is put on striking the balance between personalization and security, calling for transparency via explainable AI and privacy-sensitive user interfaces. From our analysis, we believe that adaptive cybersecurity protocols can reduce risks without significantly compromising the benefits of personalization
Management Strategies for Optimizing Security, Compliance, and Efficiency in Modern Computing Ecosystems
Get PDFThe integration of cloud, on-premises, and edge environments has increased the complexity of managing diverse computing components. This paper examines management strategies essential for maintaining efficient and resilient computing infrastructures amid rapid advancements in artificial intelligence (AI), the Internet of Things (IoT), and distributed computing. The study focuses on key areas: infrastructure management, data governance, security protocols, user access management, and resource optimization. In infrastructure management, the paper discusses hybrid and multi-cloud orchestration, load balancing, and machine learning-driven auto-scaling techniques. For data governance, it covers data lineage and metadata management platforms, data anonymization methods, and compliance automation tools to meet regulations. Security management is addressed through AI-driven threat detection using anomaly detection models, the implementation of zero-trust security architectures with micro-perimeterization, and automated incident response using Security Orchestration, Automation, and Response (SOAR) platforms. User access management strategies include policy-based access control solutions, multi-factor authentication with biometrics, and behavioral analytics. Resource optimization focuses on serverless computing models for dynamic scaling, dynamic load balancing in containerized environments, predictive resource allocation using AI analytics, and green computing practices involving dynamic voltage scaling
Drivers and Barriers of Adopting Interactive Dashboard Reporting in the Finance Sector: An Empirical Investigation
Get PDFThe finance sector has traditionally relied on static reporting methods for data analysis and presentation. With the advent of advanced data technologies, there has been a growing interest in interactive dashboard reporting. Interactive dashboards offer dynamic visualization and real-time data analysis, promising enhanced decision-making capabilities in financial contexts. Yet, the adoption of these advanced tools in the finance sector has been varied. The objective of this research was to empirically examine the drivers and barriers influencing the adoption of interactive dashboards as opposed to traditional static reporting in the finance sector. The study analyzed data collected from 381 professionals working in the finance sector, including roles such as financial analysts, data analysts, IT professionals, data engineers, finance managers, executives, and business intelligence professionals. The methodology of this study includes traditional regression methods and four machine learning algorithms: decision tree, random forest, support vector machine (SVM), and K-nearest neighbors (KNN). The target participants were categorized into three groups based on their adoption stance: not willing to adopt, undecided, and willing to adopt. Results from traditional regression methods indicated that enhanced data visualization and interactivity, real-time data analysis, and customization and flexibility positively impacted the willingness to adopt interactive dashboards. Conversely, age, cost implications, dependency on IT infrastructure and support, learning curve and training requirements, and organizational tenure were identified as significant barriers, negatively impacting adoption. Features such as improved collaboration and sharing, efficiency in reporting, scalability and integration with multiple data sources, data security and privacy concerns, cultural resistance to change, and performance issues with large datasets were found to have an insignificant impact on adoption decisions. In the machine learning analysis, SVM classification found to be the most accurate with a 93% accuracy rate, followed by decision tree (92%), random forest (91%), and KNN (90%). The most significant feature across all methods was age, consistently showing the highest importance. Other important features included organizational tenure and real-time data analysis, which were moderately important across most machine learning methods. Cultural resistance to change and dependency on IT infrastructure and support were also important in several methods. Customization and flexibility, along with enhanced data visualization and interactivity, were crucial in specific contexts, especially where data interpretation and user interaction are key. Less important features identified included learning curve and training requirements, performance issues with large datasets, and other context-specific factors such as collaboration and sharing, efficiency in reporting, scalability and integration, cost implications, and data security and privacy concerns. The findings of this study recommend the addressing of negative impacts such as age, cost, and IT dependency while utilizing positive aspects like enhanced visualization, real-time analysis, and customization to encourage the adoption of more dynamic and interactive reporting methods in the financial data analysis domain
Identification of Age Voiceprint Using Machine Learning Algorithms
Get PDFThe voice is considered a biometric trait since we can extract information from the speech signal that allows us to identify the person speaking in a specific recording. Fingerprints, iris, DNA, or speech can be used in biometric systems, with speech being the most intuitive, basic, and easy to create characteristic. Speech-based services are widely used in the banking and mobile sectors, although these services do not employ voice recognition to identify consumers. As a result, the possibility of using these services under a fake name is always there. To reduce the possibility of fraudulent identification, voice-based recognition systems must be designed. In this research, Mel Frequency Cepstral Coefficients (MFCC) characteristics were retrieved from the gathered voice samples to train five different machine learning algorithms, namely, the decision tree, random forest (RF), support vector machines (SVM), closest neighbor (k-NN), and multi-layer sensor (MLP). Accuracy, precision, recall, specificity, and F1 score were used as classification performance metrics to compare these algorithms. According to the findings of the study, the MLP approach had a high classification accuracy of 91%. In addition, it seems that RF performs better than other measurements. This finding demonstrates how these categorization algorithms may assist voice-based biometric systems
Personalized Employee Training Based on Learning Styles Using Unsupervised Machine Learning
Get PDFAdvancement in technology, artificial intelligence, and machine learning have resulted in an explosion in the creation of tech-enabled training solutions over the past decade, contributing to the popularity of personalized learning. This research advocates for a transition away from archaic, rote learning paradigms and toward individualized employee learning experiences in which instructional styles and training tactics are tailored to the requirements of each individual employee rather than standard lesson preparation that exist today. This technique can also foster enjoyable and engaging training environments that benefit both employees and the organizations. We applied unsupervised machine learning algorithm, namely, K-means, and Hierarchical clustering algorithms to classify 1000 employees into different clusters based on the Felder-Silverman Learning Styles Model (FSLSM). As expected, no one of the employees could be precisely classified into a single category, and they demonstrated a variety of learning methods and tactics. The experiments showed 3 significant clusters across the different pairs of Processing, Input, Understanding, and Perception dimensions of the FSLSM. The findings suggest that employees can be grouped into at least 3 clusters to create personalized training materials and approaches for each group. We also discussed suitable instruction techniques, contents, and paths for each cluster. The proposed model and the findings would work in both digital and offline settings.
 
INTELLIGENT PAYMENT ORCHESTRATION PLATFORMS: ASSESSING METRICS FOR EFFICIENCY, SCALABILITY, AND SYSTEM INTEROPERABILITY
Get PDFArtificial intelligence (AI)-enabled payment orchestration platforms are reshaping the global payments industry by facilitating the optimization and streamlining of payment processes. These advanced solutions contribute to greater operational efficiency, superior customer experiences, and higher revenue generation. This study delves into the core performance metrics, scalability requirements, and interoperability issues linked to AI-powered payment orchestration platforms. By addressing these pivotal dimensions, the analysis offers critical insights for businesses aiming to harness these transformative technologies and remain competitive in the rapidly evolving digital payments environment
A Comparative Analysis of Batch, Real-Time, Stream Processing, and Lambda Architecture for Modern Analytics Workloads
Get PDFThe explosion of big data has necessitated robust, scalable, and low-latency data processing paradigms to address modern analytics workloads. This paper provides a technical comparative analysis of batch processing, real-time processing, stream processing, and the hybrid Lambda architecture, highlighting their architectural principles, data flow models, performance characteristics, and trade-offs. Batch processing operates on static, large-scale datasets and prioritizes high throughput but incurs significant latency. Real-time and stream processing frameworks enable continuous or near-instant processing of unbounded data streams, focusing on minimal latency while maintaining system resilience. The Lambda architecture integrates batch and stream layers to provide fault-tolerant, scalable analytics with accurate and timely results. This paper dissects these paradigms based on technical metrics such as latency, fault tolerance, scalability, data consistency, resource utilization, and operational complexity. We further analyze real-world use cases, highlighting how each paradigm addresses specific workload requirements in domains such as IoT, finance, and big data systems. Our findings emphasize that while no single paradigm is universally optimal, selecting the right architecture requires balancing latency, throughput, and computational efficiency based on workload characteristics and business priorities