Clustering Interval-valued Data Using an Overlapped Interval Divergence

As a common problem in data clustering applications, how to identify a suitable proximity measure between data instances is still an open problem. Especially when interval-valued data is becoming more and more popular, it is expected to have a suitable distance for intervals. Existing distance measures only consider the lower and upper bounds of intervals, but overlook the overlapped area between intervals. In this paper, we introduce a novel proximity measure for intervals, called Overlapped Interval Divergence (OLID), which extends the existing distances by considering the relationship between intervals and their overlapped “area”. Furthermore, the proposed OLID measure is also incorporated into different adaptive clustering frameworks. The experiment results show that the proposed OLID is more suitable for interval data than the Hausdorff distance and the city-block distance.