A data-clustering approach based on artificial ant colonies with control of emergence

Data-clustering has been identified as a major problem in many areas. It aims to identify and extract meaningful groups from a very large set of data. It is a combinatorial problem, because the number of partitions that can be obtained grows exponentially with the volume of data to be classified and the number of clusters. In this paper, we deal with the problem from the perspective of distributed optimization and we present a new approach for data-clustering based on artificial ant colonies with control of emergence. The features of the proposed approach consist essentially in the definition of a new dynamics of ants, governed by new probabilistic rules to pick up and drop objects. A mechanism for controlling the emergence was implemented by using anti-clustering agents, and guided by the number of detected clusters. A multi-agent platform was used to implement the proposed approach. The obtained results in terms of internal and external performance measures on a set of real and synthetic benchmarks show the competitiveness of the proposed approach compared to other approaches in the literature, as well as the made modifications (contribution).