Improving classification accuracy on uncertain data by considering multiple subclasses

摘要

In this paper, we study the problem of classification on uncertain objects whose locations are uncertain and described by probability density functions (PDF). Since some existing algorithms have a bottleneck caused by expensive computational cost when handling uncertain objects, supervised Uncertain K-means (UK-means) algorithm is proposed to classify uncertain objects more efficiently. Supervised UK-means assumes that the classes are well separated. However, in real data sets, objects from the same class are usually interspersed among (disconnected by) other classes. Thus, we propose a supervised UK-means with multiple subclasses (SUMS) which considers that the objects in the same class can be further divided into several groups (subclasses) within the class. Moreover, we propose a bounded supervised UK-means with multiple subclasses (BSUMS) to avoid overfitting. We demonstrate that SUMS and BSUMS perform better than some existing algorithms by extensive experiments.