Genetic-guided semi-supervised clustering algorithm with instance-level constraints

Abstract

Semi-supervised clustering with instance-level constraints is one of the most active research topics in the areas of pattern recognition, machine learning and data mining. Several recent studies have shown that instance-level constraints can significantly increase accuracies of a variety of clustering algorithms. However, instance-level constraints may split the search space of the optimal clustering solution into pieces, thus significantly compound the difficulty of the search task. This paper explores a genetic approach to solve the problem of semi-supervised clustering with instance-level constraints. In particular, a novel semi-supervised clustering algorithm with instance-level constraints, termed as the hybrid genetic-guided semi-supervised clustering algorithm with instance-level constraints (Cop-HGA), is proposed. Cop-HGA uses a hybrid genetic algorithm to perform the search task of a high quality clustering solution that is able to draw a good balance between predefined clustering criterion and available instance-level background knowledge. The effectiveness of Cop-HGA is confirmed by experimental results on several real data sets with artificial instance-level constraints.