Enhanced Link Prediction Analysis Using Integrated Classifier

The ultimate aim of link prediction is to identify the possible potential connections in a network. The study on this topic has gained impetus as it results in efficiently saving resources, time, cost and effort to analyze future possibilities in a network. With the refined use of this technique, it significantly improves how the complex networks like social network analysis, biological networks, and recommendation systems are interpreted vis-à-vis experimental processes. In this paper, we propose a novel combination of five node centralities and four similarity measures with the aim of capturing both local and global features of networks. Consequently, feature vector made by integration of these five node centralities and four similarity indices are then passed through Machine Learning(ML) classifiers. By combination of results of different classifiers according to dynamic weighting scheme, the integrated classifier is then utilized for final link prediction. We have also analyzed the effect of varying thresholds on the ROC AUC and F1 scores and the same have been tabulated. This paper provides insights into the effectiveness of combining graph-theoretic features with ML models for accurate link prediction.