PCA-assisted clustering approaches for soft-failure detection in optical networks

Over the past years, the emergence of complex and bandwidth-hungry applications has charged the efforts to ensure the reliability of optical networks. Moreover, network scalability issues pose challenges as the number of optical parameters increases rapidly. In this regard, it is important to minimize the risk of optical failures by providing an autonomous and scalable failure detection approach. Hence, this paper presents a scalable and interpretable failure detection in optical networks exploiting five clustering algorithms (K-means, fuzzy C-means, Gaussian mixture model, DBSCAN, and mean shift) assisted by a dimensionality reduction technique. Cluster-based approaches facilitate the physical interpretability of the failure distributions among the telemetry data by allowing their clear visualization. Meanwhile, the dimensionality reduction technique can handle large-scale telemetry data with numerous optical parameters, improving the performance of clustering algorithms, as these have limitations when dealing with high-dimensional data. The proposed approaches are evaluated based on Type I/II errors (commonly known as false positive and false negative indications, respectively). A dataset derived from an optical testbed is used to evaluate the robustness of the proposed approaches.