Simultaneous Evolutionary Optimization of Features Subset and Clusters Number

Cluster analysis is a popular technique used to identify patterns in data mining. However, evaluating the accuracy of a clustering task is a challenging process which remains to be an open issue. In this work, we focus on two factors that significantly influence clustering performance: the optimal number of clusters and the subset of relevant attributes. While the former has been extensively studied, the latter has received comparatively less attention, especially in relation to its equivalent in supervised learning. Despite their clear interdependence, these factors have rarely been studied together. In this context, we propose an evolutionary algorithm that simultaneously optimizes both factors using ad-hoc variations of internal validation indices as a fitness function.