Ensemble-model-based link prediction of complex networks

The traditional similarity-based link prediction of complex networks mainly considers a certain similarity index of each node. To improve the stability and accuracy of link prediction, in this paper, we assemble the four similarity indexes (i.e. CN, LHN-II, COS+, and MFI) and introduce the idea of stacking into the link prediction of complex networks. First, based on Logistic regression algorithm and an Xgboost algorithm, we take four similarity indexes as the four characteristics to be learned by the models. Next, we use cross-validation, grid searching and early-stopping methods to determine the hyperparameters of these models. Then, according to the idea of the ensemble-model, and combined with logistic regression and stacking technology, we propose a new link prediction algorithm of complex networks, called the ensemble-model-based link prediction algorithm (EMLP). Finally, we test and validate our results by conducting numerical simulations. Using the American aviation network and nematode neural network as two real-world examples, we verify the feasibility and effectiveness of the proposed EMLP algorithm by comparing the AUC value and the Recall rate. Simulations reveal that the link predictions of the EMLP method proposed in this paper have better stability and accuracy.