Prediction of Optical Fiber Network Attenuation to Assess Network Performance Using Random Forest Regression

Abstract

The increasing demand for telecommunication services requires reliable fiber optic network infrastructure. However, signal attenuation remains a major challenge in data transmission that leads to service quality degradation. This research aims to predict fiber optic network attenuation values using the Random Forest Regression algorithm by utilizing historical data from PT. Telkom Akses Indonesia. The dataset consists of 1225 New Connection Installation (PSB) data with features including ODP coordinates, customer location, cable length, and attenuation values. Research results show that the model with a 90:10 data ratio provides optimal performance with an R² Score of 0.9832, MSE of 0.0559, RMSE of 0.2363, and MAE of 0.1009. The model successfully explains 98.32% of variation in network attenuation data. Predictions of attenuation values for future periods show a stable trend, enabling proactive assessment of fiber optic network performance so operators can anticipate disruptions before they impact customer service.