Abstract:

Unsupervised learning algorithms are used in many engineering applications since they extract information by the minimum human interaction. Modularity is one of the well known unsupervised machine learning algorithms to classify the network of information, so the relational information is highlighted. The closely related data points gather together to create relatively dense subcommunities. Thus, the meaningful relationships between data points (a.k.a. nodes or vertices) could be extended to the common properties of the data points. In this study, we aim to solve 3-KCP which is inherently a combinatorics problem by the modularity classification. Additionally, we compared the result with the well-known clustering algorithms, namely Clique Percolation, Spectral, Centrality, and Hierarchical clustering. Our investigation by means of modularity includes extended resolutions from 0.1 to 2.1, and 1.0 is the optimum resolution to find all 3-KCP solutions for the Modularity algorithm. Comparison of the modularity with the specified clustering algorithms shows the superiority of the modularity algorithm because compared algorithm provides wrong clusters by means of N-KCP or identified clusters by modularity.

Keywords: