Efficient training on biased minimax probability machine for imbalanced text classification

Xiang Peng, Irwin King

Abstract: The Biased Minimax Probability Machine (BMPM) constructs a classifier which deals with the imbalanced learning tasks. In this paper, we propose a Second Order Cone Programming (SOCP) based algorithm to train the model. We outline the theoretical derivatives of the biased classification model, and address the text classification tasks where negative training documents significantly outnumber the positive ones using the proposed strategy. We evaluated the learning scheme in comparison with traditional solutions on three different datasets. Empirical results have shown that our method is more effective and robust to handle imbalanced text classification problems.