Decomposition-Fusion for Label Distribution Learning

Label Distribution Learning (LDL) is a general learning framework that assigns an instance to distribution over set of labels rather than single label or multiple labels. Current LDL methods have proven their effectiveness in many real-life machine applications. However, generalization the classification task and as such it exposed same problems standard algorithms, including class-imbalanced, noise, overlapping irregularities. The purpose this paper mitigate these effects by using decomposition strategies. technique devised, called Decomposition-Fusion for (DF-LDL), based on one most renowned strategy decomposition: One-vs-One scheme, which we adapt be able deal with datasets. In addition, propose competent fusion method allows us discard non-competent classifiers when output probably not interest. proposed DF-LDL verified several real-world datasets carried out two types experiments. First, comparing our proposal base learners and, second, state-of-the-art algorithms. shows significant improvements both • model. classic Decomposition strategies are efficient dealing issues. apply LDL-type problems.