Unsupervised Anomaly Detection in Testchannels : A Comparison Between Machine Learning Techniques

Abstract: Due to the increasing number of mobile applications and services, communication service providers strive to optimize their networks in order to maintain a competitive position. Continuous Integration, which includes improving software delivery through automation, is fundamental in the process of testing and optimizing networks. This study aims to investigate if three different unsupervised machine learning techniques can be applied to detect anomalies in channels used for testing. The three different machine learning algorithms utilized and evaluated for this purpose are: Neural Networks, Isolation Forest, and Principal Component Analysis. The findings implies that none of the chosen models are optimal for the given task. The results are discussed in the light of previous research, and recommendations for future research are suggested.