Fiber Optical Module Anomaly Detection Using Graph Deep Learning Model

Graph deep learning models represent a novel technique in the field of machine learning. Compared to typical deep machine learning approaches, graph deep learning has the capability to store information about relationships. This is particularly useful for real-world data, especially in domains such as telecom networks, where the network architecture itself can be represented as a graph. In this study, we applied graph deep learning to real-world telecom data and analyzed it using Digital Diagnostics Monitoring (DDM) data, which monitors the status of fiber optic modules. The problem of DDM in practice often involves anomaly detection, where the patterns of anomalies can vary and cannot be directly learned from the data. To address this challenge, we employed a Graph AutoEncoder model (GAE) as a classification model to predict anomalies in fiber optic modules. Specifically, we constructed a sophisticated Graph AutoEn-coder (GAE) model that incorporates both nodes and edges within the graph. We conducted experiments using this model and evaluated its performance by adjusting internal parameters to manipulate the influence of nodes and edges. The performance metrics utilized in this study are Precision and Recall. The results indicate that incorporating information from edges can enhance the model's performance.