Comparitive Analysis of Link Prediction in Complex Networks

Abstract

The most attractive aspect of data mining is link prediction in a complex network. Link prediction is the behavior of the network link formation by predicting missed and future relationships among elements based on current observed connections. Link prediction techniques can be categorized into probabilistic, similarity, and dimension reduction based. In this paper six familiar link predictors are applied on seven different network datasets to provide directory to users. The experimental results of multiple prediction algorithms were compared and analyzed on the basis of proposed comparative link prediction model. The results revealed that Jaccard coefficient and Hub promoted performed well on most of the datasets. Different applied methods are arranged on the basis of accuracy. Moreover, the shortcomings of different techniques are also presented.