Classification Algorithm for Link Prediction Based on Generated Features of Local Similarity-Based Method

Abstract

A social network is a social structure that consists consisting of nodes, edges, or links and describes activity on a social media platform. Later, link prediction is a technique to predict new relationships for future networks based on information explored from the current network topology. Several local similarity-based methods use topological information to predict the link. However, these methods have different performances and depend on the network topology. This study proposes using classification algorithms of machine learning to predict future links. The classification algorithms compared are k-Nearest Neighbors (KNN), Naive Bayes, Decision Tree, and Random Forest by comparing six social network datasets with features generated from local similarity-based methods. This research was conducted in three stages: preprocessing, classification comparison, and performance evaluation. The findings of this study are that the Random Forest algorithm outperforms for testing accuracy, precision, and F1-Score. However, in the recall test results, Random Forest only outperformed other benchmark algorithms in the four datasets: soc-karate, soc-dolphin, soc-highschool M, and Soc-sparrowlyon-flock-season 03. Meanwhile, in the datasets soc-tribes and soc-aves-weaver-social-05, the Decision Tree algorithm outperformed other benchmark algorithms.