Evaluating and Comparing the IGraph Community Detection Algorithms

Complex networks became a very important tool in machine learning field, helping researchers to investigate and mine data. They can model real dynamic networks, aiding to unveil information´s about the systems they model. Communities are notable groups that may exist in a complex network and the community detection problem is the focus of attention of many researchers. The igraph library implements a good set of community detection algorithms, allowing researchers to easily apply them to data mining tasks. But each algorithm uses a different approach, leading to different performances. In this paper, the community detection algorithms implemented in the igraph library are investigated and ranked according to their performances in a set of different scenarios. Results show walktrap and multi-level got the highest scores while leading eigenvector and spinglass got the lowest ones. These findings are an important contribution for aiding researchers to select or discard algorithms in their own experiments using igraph library.