基于加权支持向量机的膜蛋白类型预测中不平衡问题处理

Solving the Problem of Imbalanced Dataset in the Prediction of Membrane Protein Types Based on Weighted SVM

针对膜蛋白类型预测中普遍存在的不平衡样本问题,分析一般支持向量机(SVM)在处理不平衡样本时的缺陷,引入加权SVM来补偿由于训练集中的类别差异引起的分类结果偏向于多样本类别的问题.采用统计预测中的一致测试、交叉校验和独立测试方法进行测试.实验表明,不平衡处理后的效果非常理想,该方法可以成为现有方法一个有效的补充分析工具.

To deal with the common imbalance problem in the prediction of membrane protein types, the weighted-SVM was introduced to compensate the bias toward large size training set caused by the imbalance size of imbalance training sets. The self-consistency test, jackknife test, and independent dataset test were conducted. The results indicate that the current approach may serve as a powerful complemental tool to other existing methods.