Supervised Machine Learning for Link Prediction Using Path-Based Similarity Features

Link prediction is an emerging research domain, where new ties are predicted based on the structural properties of the network. In this paper, path-based similarity measures have been adopted for link prediction which is able to extract the global similarity instead of local neighborhood similarity. It has been observed that individual similarity measures may not be a adequate for future link analysis. In this paper, several path-based similarity measures like Katz index, Adamic Adar, simRank and Hitting time measures are hybridized for feature extraction in supervised machine learning models for link prediction. The major contribution in this paper is to leverage the path-based similarity measures to construct the feature vector for supervised machine learning algorithms to predict the future link. The effectiveness of machine learning models is compared with the individual global similarity measures using a few real-world network datasets.