A New Multi-label Learning Algorithm Using Shelly Neighbors

Since multi-label data is ubiquitous in reality, a promising study in data mining is multi-label learning. Facing with the multi-label data, traditional single-label learning methods are not competent for the classification tasks. This paper proposes a new lazy learning algorithm for the multi-label classification. The characteristic of our method is that it takes both binary relevance and shelly neighbors into account. Unlike k nearest neighbors, the shelly neighbors form a shell to surround a given instance. As a result, our method not only identifies more helpful neighbors for classification, but also exempts from the perplexity of choosing an optimal value for k in the lazy learning methods. The experiments carried out on five benchmark datasets demonstrate that the proposed approach outperforms standard lazy multi-label classification in most cases.