Asymmetric Kernel Scaling for Imbalanced Data Classification

Many critical application domains present issues related to imbalanced learning - classification from imbalanced data. Using conventional techniques produces biased results, as the over-represented class dominates the learning process and tend to naturally attract predictions. As a consequence, the false negative rate may result unacceptable and the chosen classifier unusable. We propose a classification procedure based on Support Vector Machine able to effectively cope with data imbalance. Using a first step approximate solution and then a suitable kernel transformation, we enlarge asymmetrically space around the class boundary, compensating data skewness. Results show that while in case of moderate imbalance the performances are comparable to standard SVM, in case of heavily skewed data the proposed approach outperforms its competitors.