Integrating Feature Extraction with Semi-Supervised Learning for Failure Detection in Optical Networks

Since optical networks support data-intensive applications, ensuring transmission reliability is crucial. Conventional failure detection often relies on simplified thresholds or supervised learning, requiring large failure datasets. Semi-supervised methods offer viable alternatives by handling limited or imbalanced data. This work proposes a failure detection method integrating an autoregressive (AR) model and a one-class support vector machine (OCSVM). The AR model extracts relevant features, enhancing OCSVM performance. The approach was evaluated using an optical testbed dataset. Results show that integrating AR with OCSVM improved detection accuracy from 66.86% to 84.17% compared to the traditional OCSVM without AR.