부산대학교 도서관

One-class support vector machine (OCSVM) is regarded as an important one-class classification method for tackling the problem of extreme class imbalance. However, combining several OCSVMs by the traditional ensemble approaches may not improve the performance of the single OCSVM because it is known as a strong classffier. In the paper, a rotation based ensemble method is proposed for integrating OCSVMs., The training data of different OCSVMs in the ensemble is transformed by different rotation matrices. Therefore, the diversity of training data for the ensemble of OCSVMs can be guaranteed. Experiments conducted on ten benchmark data sets to validate the effectiveness of the proposed ensemble strategy.