Abstract

Advanced fault management techniques beyond conventional reactive procedures are required due to the widespread use of Gigabit Passive Optical Networks (GPON) as essential infrastructure for high-speed internet services.  In order to anticipate connectivity problems before service degradation happens, this study introduces a revolutionary proactive fault detection and monitoring system that combines machine learning algorithms with real-time network analytics. Our hybrid technique addresses class imbalance issues while preserving real-world representativeness by combining meticulously vetted synthetic samples with real failure data from Telekom Malaysia's operational GPON infrastructure. In order to forecast five different fault categories—Line Disconnect, Intermittent Failures, Service Down, Frequent Disconnections, and Normal Operation—the system examines crucial network data such as optical power levels, signal-to-noise ratio, reflectance measures, and signal attenuation. Our Support Vector Machine solution achieved 97% classification accuracy with balanced precision and recall across all fault types after thorough evaluation utilizing several machine learning methods. During a six-month operational trial, the implementation of a web-based monitoring dashboard showed practical success with a mean time to fault resolution reduction of almost 60%. Crucially, this study clearly defines the parameters for model generalizability across various network topologies and operating situations and offers an open discussion of the constraints of synthetic data.

This work is licensed under a Creative Commons Attribution 4.0 License.