On Indicator Functions for Evaluating Naturality of Clustering Results
Tomotake Nakamura · 2007
Our aim is to realize user-friendly clustering for large high-dimensional data sets. It is dicult to choose a desired clustering result among natural ones. In order to obtain a desired clustering result for large high-dimensional data sets, we found that it is necessary to perform interactive clustering. We have already pro- posed an indicator function for choosing a more natural clustering result than the other one. In this paper, we propose a new indicator function based on the degree of outliers, which chooses more natural clustering result than one chosen by the previous indicator function. In simulation experiments with benchmark data, we show eectiveness of the new indicator function.