Extension Classification Method for Label Variability

Traditional classification algorithms typically assume that the labels of training samples are static and deterministic, ignoring the dynamic characteristics of sample labels that may change with conditions in real-world scenarios. In response to this issue, this paper proposes a new learning problem setting—the Extended Classification Problem, which simultaneously gives the class labels and label variability states of samples in the training data, which to characterize the class transition potential of samples under the influence of change mechanisms. Based on this setting, a multi-label learning framework was designed, an extension classification algorithm for label variability using support vector machine was constructed, to achieve collaborative optimization of category discrimination and label variability prediction. The experimental section validated the effectiveness of the proposed algorithm on both synthetic and real datasets. This paper provides a new modeling approach for label dynamic learning problems, which has good application prospects.