Análise comparativa entre algoritmos de agrupamento e de detecção de comunidades em redes

Abstract

Clustering is one of the most notorious Machine Learning techniques and has an infinite number of practical applications in different areas of knowledge. Understanding the types of approaches and their characteristics is essential for successful clustering. In this context, in addition to traditional clustering algorithms such as DBSCAN, an approach that has gained notoriety in the literature is network-based clustering algorithms, which build a network from data and use communities to find groups. The purpose of this study is to carry out a comparative analysis between clustering algorithms and community detection algorithms in experiments performed with an artificially generated database. In addition, an analysis of the impact of graph generation algorithms on the performance of community detection algorithms was performed. For the results, the Normalized Mutual Information (NMI) was calculated, one of the main metrics in the performance evaluation in clustering problems. In the results of the experiments it was analyzed wheter, even with a difference in the algorithm performance, such difference represents or not a statistical difference. The results showed that community detection algorithms are a viable alternative and can outperform traditional clustering algorithms in some scenarios.