Link Prediction in Brain Networks with Epilepsy Disease

Abstract

Node similarity refers to the link probability in complex network. Higher value signifies their strong possibilities for interlinkages. In this paper, four techniques were tested for node similarity evaluation including CN (common neighbor), RA (resource allocation), AA (Adamic-Adar) and Sorenson algorithm on several real and simulated networks. AUC (Area Under the Curve) was set to compare their prediction accuracy. It's found out that RA performs much better and then it was adopted on frequency dependent brain networks with generalized lobe epilepsy disease. Similarity matrix was produced and the results demonstrated that RA could help in uncovering pathological mechanism through leading nodes identification based on node similarity.