Data Selection for Exact Value Acquisition to Improve Uncertain Clustering

Abstract

In recent years, data uncertainty widely attracts researchers’ attention because the amount of imprecise data is growing rapidly. Although data are not known exactly, probability distributions or expected errors are sometimes available. While most researchers on uncertain data mining are looking for methods to extract mining results from uncertain data, which is usually in the form of probability distributions or expected errors, it is also very important to lower the data uncertainty by making a part of data more certain to help get better mining results. For example, input values of some sensors in the sensor network are usually designed to be recorded more frequently than others because they are more important or more likely to change. In this paper, the issue of selecting a part of uncertain data and acquiring their exact values to improve clustering results is explored. Under a general uncertainty model, we propose both global and localized data selection methods, which can be used together with any existing uncertain clustering algorithm. Experimental results show that the quality of clustering improves after the selective exact value acquisition is applied.