Detecting Nested Structures Through Evolutionary Multi-objective Clustering.

The evolutionary multi-objective algorithms have been widely applied for clustering. However, in general, the detection of heterogeneous nested clusters remains challenging for clustering algorithms. This paper proposes an adaptation of the connectedness criterion used as an objective function in established Evolutionary Multi-Objective Clustering approaches (EMOCs). This adaptation can improve the conflict between the objective functions, and then it promotes the detection of nested clusters. We performed experiments with four EMOCs (MOCK, MOCLE, Delta-MOCK, and EMO-KC) that provide different features. These different EMOCs have different initialization methods and representation schemes, allowing us to analyze how the proposed objective function can contribute to detecting nested clusters. Our results show that our adapted objective function promotes a general gain in the performance of all these algorithms.