Oversampling multi-label data based on natural neighbor and label correlation

Bin Liu, Ao Zhou, Bingkun Wei, Jin Wang, Grigorios Tsoumakas

Abstract: Highlights•Retrieving adaptive and solid neighborhood for multi-label data.•Emphasizing boundary areas of highly imbalanced labels.•Exploiting label correlations to guide the label assignment.•Better performances than 8 baselines with 5 multi-label classifiers.