Predicting the performance of broadband passive optical networks using machine learning

Abstract Broadband passive optical networks is an established architecture for the high-speed data transfer. For effective fault diagnosis and self-configuration in these networks, analysis of network-generated data is essentially required. In this era, machine learning-based data analytics approaches could play a vital role in analyzing the performance of the networks. In this paper, a machine learning approach has been proposed for predicting the performance of broadband passive optical networks. For this task, a dataset consisting of fiber length, transmission power, number of power splitters, line width, and extinction ratio parameters has been generated to make an estimate of the Q factor for a given optical network. Out of these network parameters, fiber length, transmission power, and several power splitters are selected through the relief attribute evaluation technique. The selected parameters are fed into a regression-based model tree classification algorithm for estimating different levels of Q factor. This work also takes into account logistic regression, decision tree, decision table, PART, and random forest algorithms for the desired task. The analysis of simulation results proves that the regression-based model tree classification algorithm provides an effective estimate of Q factor in terms of accuracy of 93.23% and 95.74% for 7-class and 3-class problems. Thus, this algorithm appears to be a suitable choice to predict the performance of broadband passive optical networks accurately.