I Interval Set Representations of Clusters

The amount of information that is available in the new information age has made it necessary to consider various summarization techniques. Classification, clustering, and association are three important data-mining features. Association is concerned with finding the likelihood of co-occurrence of two different concepts. For example, the likelihood of a banana purchase given that a shopper has bought a cake. Classification and clustering both involve categorization of objects. Classification processes a previously known categorization of objects from a training sample so that it can be applied to other objects whose categorization is unknown. This process is called supervised learning. Clustering groups objects with similar characteristics. As opposed to classification, the grouping process in clustering is unsupervised. The actual categorization of objects, even for a sample, is unknown. Clustering is an important step in establishing object profiles. Clustering in data mining faces several additional challenges compared to conventional clustering applications (Krishnapuram, Joshi, Nasraoui, & Yi, 2001). The clusters tend to have fuzzy boundaries. There is a likelihood that an object may be a candidate for more than one cluster. Krishnapuram et al. argued that clustering operations in data mining involve modeling an unknown number of overlapping sets. This article describes fuzzy and interval set clustering as alternatives to conventional crisp clustering.