An Evaluation of Various Regression Models for the Prediction of Two-Terminal Network Reliability

Abstract

Analyzing network data is presently a big challenge for applied machine learning. Many model architectures have been proposed to study or extract information from network data for specific applications. In this paper, we compare the performance of autoencoders, convolutional neural networks and extreme gradient boosting decision trees with different configurations for the task of approximating two-terminal network reliability. The ground truth is generated using an analytical method. Various synthetic datasets containing networks with different configurations are used. The obtained results help us to identify the dataset factors which affect the prediction performance of these models.