Anomaly Detection in Network Monitoring : a comparison of performance analysis of generic vs cloud-based Machine learning Platform
Nukhba, Nukhba (2023)
Nukhba, Nukhba
2023
Master's Programme in Computing Sciences
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2023-06-12
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202306086630
https://urn.fi/URN:NBN:fi:tuni-202306086630
Tiivistelmä
Applications of anomaly detection in the telecommunication industry are widely in demand for the purpose of automating network supervision and providing better network reliability at a reduced cost. To meet this increasing demand, various machine learning platforms have emerged that empower businesses to leverage ML capabilities at a better cost. This study compares a generic machine learning platform sci-kit-learn with Azure Machine learning studio, while taking anomaly detection as a center use case. The focus of this thesis is to understand the value offered by the Cloud giants for the application of anomaly detection and possibly suggest a better platform for the development of machine learning applications, deployment, and maintenance.
This thesis studies a generic and cloud-based machine-learning platform by employing various machine-learning methods for anomaly detection. The machine learning methods from a generic platform and a cloud-based platform are applied to two anomalous labeled network datasets. These platforms are analyzed with respect to data processing, model creation, and prediction, cost, and usability.
As a result, this work concludes the comparison with various aspects of model creation, visualization of the results, model performance, cost, and ease of use. Results show that cloud-based machine learning platforms do provide an edge with respect to usability by providing low to no code options. However, low code also implies less control over the model, and it also comes with limited ready-to-use algorithms. And even less built-in support for unsupervised algorithms in Azure Designer. The whole process of model creation, data processing, and deployment was effortless, thus, it is easy to convert to production ready.
In addition, cloud-based platforms eliminate the need of buying high-end machines for storage computation. However, the studio charges money per experimentation hour and storage. Both platforms are well suited for the purpose of anomaly detection with each having some advantage over the other.
This thesis studies a generic and cloud-based machine-learning platform by employing various machine-learning methods for anomaly detection. The machine learning methods from a generic platform and a cloud-based platform are applied to two anomalous labeled network datasets. These platforms are analyzed with respect to data processing, model creation, and prediction, cost, and usability.
As a result, this work concludes the comparison with various aspects of model creation, visualization of the results, model performance, cost, and ease of use. Results show that cloud-based machine learning platforms do provide an edge with respect to usability by providing low to no code options. However, low code also implies less control over the model, and it also comes with limited ready-to-use algorithms. And even less built-in support for unsupervised algorithms in Azure Designer. The whole process of model creation, data processing, and deployment was effortless, thus, it is easy to convert to production ready.
In addition, cloud-based platforms eliminate the need of buying high-end machines for storage computation. However, the studio charges money per experimentation hour and storage. Both platforms are well suited for the purpose of anomaly detection with each having some advantage over the other.