Network traffic anomaly detection by dimensionality reduction methods

Anomaly in network traffic can arise from various cyberattacks, malfunction network devices or software flaws. Therefore traffic anomaly detection is an important practical problem especially in the issue of cybersecurity. Many recent anomaly detection techniques are based on machine learning methods. Unsupervised machine learning algorithms generally are more preferable than supervised learning because they can be trained on raw undimensioned data that greatly reduce training expenses. Dimensionality reduction methods are unsupervised learning algorithms that can be used for traffic anomaly detection. This work presents an application of such traffic anomaly detection methods, when faults are caused by some network attacks.