PCA-Assisted Fuzzy Clustering Approach for Soft-Failure Detection in Optical Networks

This paper presents a scalable and interpretable failure detection in transparent optical networks (TONs) exploiting the combination of a Fuzzy C-means clustering approach assisted by a Principal Component Analysis (PCA) based dimensionality reduction technique. As cluster-based approaches allow clear visualization of the data, the use of such techniques can improve the interpretability of the failure detection performance. Meanwhile, the dimensionality reduction technique can handle the typical large-scale telemetry data collected in real-world environments. Evaluations in an optical network testbed show the effectiveness of optical failure detection in terms of classification errors.