Semi-supervised Classification and Noise Detection

Semi-supervised learning has become a topic of significant interests recently. In this paper, we are concerned with semi-supervised classification and noise detection. Based on label propagation algorithm, we present an improved label propagation algorithm, which can classify data and detect noise simultaneously. Compared with original label propagation algorithm, by detecting noise and constraining some labels that can be propagated, the improved algorithm can prevent propagating mislabels and avoid results´ tendency to the larger number of labels, so as to improve the semi-supervised classification results. Experimental results demonstrate the effectiveness of this algorithm.