A class of dynamic rough partitive algorithms

Abstract

Partitive algorithms, like cluster algorithms, are frequently used methods in data mining. Most of them are static in the sense that they detect pattern in stable data structures, i.e. the data structure remains unchanged over time. However, many real-life situations are characterized by changing data environments that require an adaptation of the respective algorithms. The retail industry, for example, is frequently faced with changes in its customers' buying behavior. Segments of customers that are buying certain goods may vary within one year or even within one single day. Detecting these changes could provide business opportunities for the respective retailer. For such changing data structures, dynamic clustering algorithms have been developed that adapt to changing environments. This paper discusses a class of rough partitive algorithms and introduces their possible dynamic extensions. C 2011 Wiley Periodicals, Inc.

Dynamic clustering with soft computing

“…In particular, partitive clustering algorithms consist of three steps 6. These principle steps of cluster analysis (see Figure 1) are the definition of the initial settings, in particular, the number of clusters, classifier design, and the optional last step classification (e.g., note that in machine learning, the term ‘classification’ is often used in a different meaning—interchangeable with supervised learning contrasting clustering as an unsupervised learning method).…”