Effective Link Prediction in Complex Networks Using Differential Evolution Based Extreme Gradient Boosting Algorithm

Abstract

Link Prediction is the estimation of missing links or possible future links between any two nodes in a network. Based on the type of network, the use cases for these models may vary. Recommendation systems in eCommerce websites, relationship prediction in social media networks, and biological gene link assessment in the Human body are a few examples of the application of Link Prediction. The conversion of this problem into a supervised classification model is expected to yield reliable results. Ensemble learning models such as XG – Boost (XGB) combine multiple small decision trees to learn the input data and produce accurate results. But the results produced by the model vary highly based on the hyperparameters of the XGB. With the help of Differential Evolution (DE), these parameters can be optimized to perfectly fit the dataset. This paper attempts to solve the link prediction problem in complex networks using Differential Evolution-based XGB (DXGB). The working of DXGB is tested extensively on various real-world complex networks and is showcased. The experimental results reveal that the usage of DE with XGB produces a better prediction of links in complex networks. The implementation along with the results obtained is presented in this paper.