8333 research outputs found
Sort by
Advanced Visa Outcome Predictions for Superior Accuracy and Interpretability
No full textThe H1B visa plays an important role for skilled workers looking for employment in the U.S.; however, its application process is unclear and inconsistent, presenting challenges for employers. This study addresses these issues using two robust predictive models: the Bi-LSTM model for sequential data and XGBoost for structured data analysis to predict H1B visa outcomes with high accuracy and interpretability. This research applies advanced feature selection and data balancing methods to H1B visa data from the 2017 to 2022 fiscal years to address class imbalances and achieve highly generalized models. The deep learning and machine learning models are employed to find a factor influencing visa decisions. More complex sequential dependencies are generated with the help of Bi-LSTM, while enhanced scalability and interpretability are derived from XGBoost. As evaluation measures, accuracy, F1 score, and recall were adopted. These metrics show improved forecasting and efficiency, along with greater transparency in the decision-making process. It offers practical recommendations for applicants and immigration authorities while offering a starting point for applying predictive modeling to additional concept classes. To eliminate unfair practices in the issuance of visas, the study aims at making the process more transparent and efficient
A Generative AI Framework for Data Augmentation Employing Generative Adversarial Networks to Predict Parkinson’s Disease
No full textEarly detection of Parkinson’s disease (PD), a degenerative neurological ailment, can lead to more effective treatment. Traditional diagnostic testing methods often rely on clinical observations analysis, which might delay diagnosis. Voice sample analysis has recently been found to be a possible early sign of Parkinson’s disease (PD) because vocal problems are linked to motor weakness. Deep learning (DL) algorithms and other advanced AI based prediction models are not as useful as they could be because there are not enough big datasets that have been labeled. This study looks into whether synthetic data generated with a Generative Adversarial Network (GAN) can increase the accuracy of Parkinson’s disease prediction. We conducted experiments and generated multiple different versions based on the number of voice recordings of PD patients and evaluated the quality of synthetic data by predicting the PD using several machine and deep learning classifiers, including random forest, XGBoost, artificial neural networks (ANN), convolutional neural networks (CNN), fully connected neural networks (FCNN). The results show that using GAN-generated synthetic data improved diagnostic performance across many models, specifically deep learning models, where ANN and FCNN achieved the 99% of accuracy rate in predicting PD for all the synthetic data versions compared to 89% for original data, representing a significant 10% increase in this study. For machine learning models the precision, recall, and f1-score values were all around 98% in all the versions of synthetic data. These results also underline the need of generative artificial intelligence in improving medical diagnosis and point to possible uses in healthcare industry
PhaseNet and EfficientNet-B0 for Phase Detection and Arrival Time Prediction
No full textSeismic phase detection and arrival time prediction are crucial for earthquake monitoring and early warning systems. This study evaluates the performance of two advanced deep learning models, PhaseNet and EfficientNet-B0, on the INSTANCE dataset for phase detection and the picking of P and S waves using spectrogram and waveform data. PhaseNet achieved a testing accuracy of 94.7%, demonstrating its effectiveness in classifying seismic events. Conversely, EfficientNet-B0 excelled in arrival time prediction, achieving a P-wave MAE of 279 ms and an S-wave MAE of 255 ms, surpassing PhaseNet in regression tasks. While PhaseNet exhibited faster training convergence, EfficientNet-B0 delivered superior accuracy and generalization. This paper highlights the complementary strengths of PhaseNet and EfficientNet-B0 in seismic phase detection and arrival time picking, contributing to advancements in seismic monitoring methodologies
Improving Fake Review Detection in E-commerce using Combined Analysis Techniques
No full textThe problem of fake reviews has become widespread and is a constant threat to most e-commerce platforms impacting the consumer and business world. The research question of this study is as follows, “How can detecting fake reviews in e-commerce be improved using combined analysis techniques?” The dataset used in this study is obtained from Kaggle; this comprises numeric features like ratings, helpfulness votes, and polarity scores from a sentiment analysis of the text content. Two modeling scenarios were explored: The focus on imbalanced data and the outcome of numeric features on balanced data. The use of techniques to resample the dataset led to an overwhelming increase in minority class detection (unverified reviews). The Random Forest/Decision Tree model had 94% accuracy and the Gradient boosting/ XGBoost model had 87% and 94% accuracy respectively. However, there are limitations to the generalization of fake reviews where the approach is reported to have low precision. The study confirms that numeric and text features are promising for fake review identification and underscores the future directions in feature selection, feature combination as well as algorithm fine-tuning. These findings offer implications for Theory development and practical use in highlighting the significance of the proper anti-fraud mechanisms in e-commerce structures
Training Investments and Innovation Gains in Knowledge Intensive Businesses: The Role of Firm Level Human Capital and Knowledge Sharing Climate
No full textTraining investments are important in securing innovation gains. However, research on this relationship in knowledge intensive businesses is nascent. In particular, questions remain concerning what value different types of training hold for different types of innovation, and what mechanisms underpin these relationships. Drawing on human capital resources theory and collective learning theory, we develop and test a model explicating how specific and general training investments, through firm level human capital, lead to incremental and radical innovation. Additionally, we propose and investigate the supposition that the predicted positive relationships between training investments, firm level human capital, and innovation will be stronger when knowledge sharing climate is high. We test our model with two-wave, multi-respondent panel data gathered from 816 knowledge intensive businesses in France, Finland, Sweden, and the UK. We find that specific training is positively related to incremental innovation but not radical innovation, whereas general training is positively related to both types of innovation. With respect to firm level human capital, we find that it mediates these relationships and they are stronger when knowledge sharing climate is high. Furthermore, our analysis reveals that knowledge sharing climate moderates both the relationship between the two types of training investments examined and firm level human capital, and the indirect relationship via firm level human capital to incremental and radical innovation. We discuss the implications for theory, research, and practice
Enhanced Automation Solution for Multi-Cloud platform: Leveraging Advanced CI/CD Tools for Deployment, Security and Testing
No full textThis research establishes an innovative technique to solve critical challenges of DevOps Practices which includes vendor lock-in, deployment complexity, Integration of automated testing in multi-cloud and security issues in multi-cloud. Most of the previous research has explored the use of different tools like Terraform and Docker automating infrastructure management, deployment optimization and testing problems. However, many of these studies are restricted to single-cloud environments, fail to address the problems of vendor lock-in and in many cases overlook the critical phases of testing, security and deployment phases of CI/Cd pipelines in multi-cloud. The research often focuses on isolated tool comparisons such as Terraform Vs Pulumi AWS-specific solutions or other cloud provider solution without thinking about complex multi-cloud deployment, interoperability issues, testing issues and security challenges posed by using different cloud infrastructures. This research automated DevOps practices in multi-cloud by enhancing the integration of Terraform, Jenkins, GitHub and Docker using different techniques that solve these gaps and enable dynamic workload migration and cross-cloud orchestration. It leverages the terraform tools to use infrastructure as code to handle the infrastructure in multi-cloud and Docker to containerise the applications and Jenkins plays an important role in this which is used to automate the process of CI/CD. It automates dynamic deployments and allows continuous integration and delivery across multiple Cloud Providers. It also enhances the testing and security enforcement with the CI/CD pipeline. Automated testing and deployment, containerized applications and security policies is seamlessly integrated into the Jenkins pipeline with other tools which ensure compliance and operational standards across multi-cloud environments. This research provides a comprehensive solution which integrates different tools terraform, Docker, and Jenkins to address the issues in multi-cloud Environments providing enterprise solutions for the application with scalable, secure and cloud agnostic for CI/CD and infrastructure management. The system achieved 0% error rates across all test cases, with AWS handling a throughput of up to 118.41 hits/second and an average response time of 318.3 ms under heavy traffic, demonstrating its robustness in managing high-traffic and write-intensive workloads. Google Cloud, on the other hand, managed a higher throughput of 152.6 hits/second with an average response time of 246.68 ms, showcasing its efficiency and cost-effectiveness for dynamic scaling and rapid deployment. While AWS is optimal for enterprise-level applications requiring high reliability and performance under complex workloads, Google Cloud is better suited for agile projects and smaller workloads, emphasizing cost efficiency and quick deployments. Both platforms displayed excellent scalability and operational reliability across varying traffic conditions
Cloud Resource Management using SLA parameters with RFD Algorithm
No full textIn the fast-evolving area of cloud computing, cloud resource management is a significant challenge faced during resource allocation and meeting of Service Level Agreements (SLAs) is crucial to maintaining users’ trust and delivering high-quality services. Most of the existing methodologies focus on execution time and cost optimization to the neglect of SLA compliance. In this work, a novel multi objective optimization framework is proposed with an improved River Formation Dynamics (RFD) algorithm with dynamic SLA parameters, machine learning (ML) based workload classification and energy-aware allocation for cloud resource management. The classified system integrates ML based workload classification to achieve 92-100% classification accuracy with respect to the different types of workloads tested. An optimal balance of SLA compliance, cost effectiveness, and energy efficiency is achieved through a weighted objective function. The simulation performance is then evaluated through experimental analysis in CloudSim which simulates 3 heterogeneous hosts and 5 VMs processing 100 cloudlets achieving 70% SLA compliance over all workload types. With the power consumption ranging between 120W–12.990kW, the system achieves optimal resource utilization of 85.2% CPU utilization for compute intensive tasks. The workloads too were distributed balanced among CPU-intensive (34%), memory-intensive (33%) and I/O intensive (33%) tasks and the execution pattern of the jobs is predictable with I/O intensive jobs completing first (1165.06s), followed by memory and CPU intensive jobs (2131.36s and 2793.87s, respectively)
An Active Defense Security Framework using Deep Reinforcement Learning for Container-Based Architectures
No full textIn this research, we present an active defense security framework for the container environments for overcoming the dynamic security challenges of containerized applications. We employ deep reinforcement learning (DRL) to optimize adaptive threat response and resource utilization within the framework. It presents a dynamic Holistic System Attack Graph (HSAG) model of time varying behavioral analysis of container activities, along with a novel Prioritized Dueling Double Deep Q Network (P3DQN) for security optimization at its foundation. A comprehensive security evaluation system is designed in the framework that is capable of real time monitoring, analysis and automatic response. Detection rates reported are 98.5% for CPU attacks and 96.8% for memory incident with 0.8% false positive rate. During normal operation, average CPU usage never exceeded 3.19%, and peak usage during active incident response was 25.04%. For detected incidents, the framework achieved an average of 17.42 seconds from detection to completion of the action (17.42s onset to action complete with a prevention rate of 100 per cent. This shows the effectiveness of the framework to provide real time security monitoring and response with optimal resource utilization in containerized service environments
Vehicle Number Plate Recognition and the Slot Allocation in a Cloud-Based Automatic Parking Management System under varying Weather Conditions
No full textRise in the growth parking issues has been caused by the increase in urbanization and the use of cars, most of which clog roads, causing traffic jams, pollution, and time exhaustion. As a solution to the issues mentioned above, we have designed an Intelligent Parking Solution based on IoT, ANPR, and Computer Vision. This system applies real-time data, artificial intelligence, and navigated cloud services to optimize parking organization, minimize parking time, and advance environmentally friendly urban mobility. The key features which includes the real-time parking slot availability, management of entry and exit through number plate recognition, automatic parking slot assignment, automated fee collection, and a centralized database for analytics and business intelligence data. The solution raises user satisfaction, increases income, optimizes flow, and supports urban growth. The YOLO-based models used for parking detection and car number identification allow the system to work under different environmental conditions. Besides these advantages, the system provides an easy-to-use interface for users, promotes the environmental conservation and helps in the development of towns and cities. By enhancing parking management, it supports smart city strategies for improved sustainable urban mobility
Analysis of hybrid encryption in cloud environments for privatization of health data
No full textTimes have revolutionized the cloud domain with its data storing capabilities and hosting services. This incentivizes the concept of remote storage for total user accessibility. However, this data is not secured and is put to risk for data misuse. This often leads to exposure of user data and breaches the integrity of the system. Moreover, since cloud is an open-source platform; simultaneous services to users often leads to data leaks. To avoid such scenarios; it has become crucial to secure data repositories from third party usage to be protected on the cloud. Therefore, the proposed research thesis aims to maintain the security and confidentiality of the system while deploying a healthcare app on various cloud platforms and leverage Cloud Service Providers(CSPs). A hybrid encryption mechanism is created from a novel approach using Advanced Encryption Standards(AES) and Advanced Encryption with Associated Data (AEAD). Additionally, Amazon Web Services(AWS) along with Continuous Integration and Continuous Delivery(CI/CD) Pipeline is also integrated into the project to ensure security and data integrity such as CodeRun, CodeDeploy and CodeBuild. The time metrics of the text conversion is evaluated, furthermore through case studies and Avalanche score is generated to indicate the strength of the algorithm with results reaching up to 93.8%. MySQL is used in the back end to manage the database and monitor user uploads. The healthcare app is finally deployed on cloud using the multi cloud technique using Microsoft Azure and Elastic Beanstalk