Soft constraints based clustering

Clustering is an important human activity. Object groups sharing the same characteristics have a significant role in human perception of the world. Clustering has many application areas. Real world applications demand for cluster analysis which satisfies various user/domain constraints. But, it becomes an important challenge to obtain meaningful clusters by solely tuning clustering parameters, especially when high dimensional and/or high volume data sets are considered. As a result, in many of such applications, user preferences and domain constraints should be taken into consideration. The objective with this work is to obtain disjoint partial clusterings by employing soft constraints. Fuzzy, probabilistic and weighted semi-rings are used to do the clustering and as well to assess the degree of soft constraints satisfaction. Genetic algorithms are used for optimization purposes. A tool, written in Java, is developed to implement what is considered. The tool has the capability of accepting/exploiting user defined soft constraints, executing clustering algorithms, displaying data sets and resulting clusterings, and calculating the clustering metrics and validity indices. The tool is experimentally evaluated on select datasets to obtain soft constraints based clusterings. To assess the performance, extensive experimental results are presented and analyzed.