Predicting Percolation Threshold of Networks using Machine Learning

Patwardhan, Siddharth (2019) Predicting Percolation Threshold of Networks using Machine Learning. Masters thesis, Indian Institute of Science Education and Research Kolkata.

PDF (MS dissertation of Siddharth Patwardhan (14MS080)) 14MS080.pdf - Submitted Version Restricted to Repository staff only Download (761kB)

Abstract

The Percolation Threshold of a network is an essential estimator of the robustness of a network against random failures. However, the computation of the percolation threshold of large networks is a computationally expensive process and therefore it is important to have methods, that do not rely on numerical simulation to predict percolation threshold of networks. This thesis focuses on the application of various machine learning based regression techniques to predict the value of the percolation threshold of networks. The data-set generated to train the machine learning models consists of various statistical and structural properties of networks as features and the numerically estimated perco- lation threshold as the output attribute. The data-set contains a total of 2000 real and synthetic networks. The performance of various machine learning models in predicting the percolation threshold has been compared and it was found that the shallow neural network model outperforms the existing estimates of percolation threshold significantly.

Actions (login required)

View Item