Dimensionality Reduction for Supervised Learning in Link Prediction Problems

In recent years, a considerable amount of attention has been devoted to research on complex networks and their properties. Collaborative environments, social networks and recommender systems are popular examples of complex networks that emerged recently and are object of interest in academy and industry. Many studies model complex networks as graphs and tackle the link prediction problem, one major open question in network evolution. It consists in predicting the likelihood of an association between two not interconnected nodes in a graph to appear. One of the approaches to such problem is based on binary classification supervised learning. Although the curse of dimensionality is a historical obstacle in machine learning, little effort has been applied to deal with it in the link prediction scenario. So, this paper evaluates the effects of dimensionality reduction as a preprocessing stage to the binary classifier construction in link prediction applications. Two dimensionality reduction strategies are experimented: Principal Component Analysis (PCA) and Forward Feature Selection (FFS). The results of experiments with three different datasets and four traditional machine learning algorithms show that dimensionality reduction with PCA and FFS can improve model precision in this kind of problem.