Link Prediction Model Based on the Topological Feature Learning for Complex Networks

Link prediction tremendously gained concern in the field of machine learning by virtue of its real-world applicability on various fields including social network analysis, biomedicine, e-commerce, criminal activities, scientific community, etc. Several link prediction methods exist which are applicable to specific types of networks. Here, the primary aim of this paper is to perform feature extraction from the given real-time complex network using subgraph extraction technique and labeling of the vertices in the subgraph according to the distance from the vertex associated with each target link. The vertices in the subgraph are labeled based on the Geometric mean distance and Arithmetic mean distance. This proposed model helps to learn the topological pattern from the extracted subgraph. The feature extraction is carried out with different size of the subgraph with the number of vertices as K = 10 and K = 15. These features are then fit into different machine learning classification models and deep learning convolutional neural network model. For the evaluation purpose, area under the receiver operating characteristic curve (AUC) metric is used. The AUC results obtained from all the classifiers have been shown. Further, the simulation results show that bagging and random forest achieved good performance. Finally, the comparative study is performed to summarize the results and proved that link prediction using classification models and deep learning model perform well across different kinds of complex networks. This solved the link prediction problem with superior performance and with robustness.