Clustering qualitative data based on the flow networks

Abstract

The paper presents research results referring to the use of flow networks and ant colony algorithm in the problem of generating decision rules for the cluster analysis process. The experiments showed that proposed approach may prove particularly important, when we are dealing with data sets represented by categorical variables associated with the same number of objects for each variance. There are many cases when we have no knowledge about the allocation group of individual data received and in addition defining any metric to measure the distance between observations, does not give any satisfactory results. Meanwhile, the selection of features and the choice of the performance metric is the basic condition for the use of most known classifiers. The article presents a new approach to solve this problem and obtain satisfactory results. It is based on mapping the set of analyzed data into the flow network, calculating the maximum flow and determining the validity of nodes in the network to use the Ant Colony Algorithm to structuring information and determine significant relationships between data.