A novel oversampling technique based on the manifold distance for class imbalance learning

Oversampling is a popular problem-solver for class imbalance learning by generating more minority samples to balance the dataset size of different classes. However, resampling in original space ineffective datasets with overlapping or small disjunction. Based on this, novel oversampling technique based manifold distance proposed, which new sample produced terms distances among neighbours space, rather than Euclidean them. After mapping data its structure, overlapped majority and will lie areas easily being partitioned. In addition, are generated locating nearby avoiding adverse effect disjoint Following that, an adaptive adjustment method presented determine number newly according distribution density matched-pair data. The experimental results 48 imbalanced indicate that proposed has better classification accuracy.