K-Nearest Neighbor Classifier for Uncertain Data in Feature Space

Uncertain data, where each feature is represented by probability density functions instead of fixed values, have been widely used in diverse applications such as sensor networks, medical data, and semiconductor wafer data. This paper proposes a new kernel function based uncertain K-nearest neighbor classifier (uncertain K-NN) algorithm for uncertain data objects in feature space. Assuming normality in the feature space, we utilize a kernel Bhattacharyya probabilistic distance measure for probabilistic distance measures. We compare the proposed uncertain K-NN classifier in feature space to an existing classifier, namely, the K-Nearest Neighbor classifier in the original space. The experimental results show the advantages of the proposed classifiers with both simulated and real data.