Data-dependent interval partition of naturally ordered individuals by complete cluster analysis in epidemiological and cardiac data processing

Cluster analysis cannot usually examine every possible clustering. However, some external constraint may reduce the number of clusterings to a practical amount. In this paper, we consider naturally ordered individuals and calculate the number of interval partitions among them, if a minimal width for each interval is demanded. Two examples illustrate the application of the method. We first determine optimal information preserving age intervals due to the frequencies of some diseases. In the second example, we find a subinterval of the intraventricular blood pressure curve suitable to determine the time constant of heart relaxation. © 1997 John Wiley & Sons, Ltd.