Strategies for the increased robustness of ant-based clustering

This paper introduces a set of algorithmic modifications that improve the partitioning results obtained with ant-based clustering. Moreover, general parameter settings and a self-adaptation scheme are devised, which afford the algorithm's robust performance across varying data sets. We study the sensitivity of the resulting algorithm with respect to two distinct, and generally important, features of data sets: (i) unequal-sized clusters and (ii) overlapping clusters. Results are compared to those obtained using k-means, one-dimensional self-organising maps, and average-link agglomerative clustering. The impressive capacity of ant-based clustering to automatically identify the number of clusters in the data is additionally underlined by comparing its performance to that of the Gap statistic. © Springer-Verlag Berlin Heidelberg 2004.