不平衡数据集中的组合分类算法

为提高少数类的分类性能,对基于数据预处理的组合分类器算法进行了研究.利用Tomek links对数据集进行预处理;把新数据集里的多数类样本按照不平衡比拆分为多个子集,每个子集和少数类样本合并成新子集;用最小二乘支持向量机对每个新子集进行训练,把训练后的各个子分类器组合为一个分类系统,新的测试样本的类别将由这个分类系统投票表决.数据试验结果表明,该算法在多数类和少数类的分类性能方面,都优于最小二乘支持向量机过抽样方法和欠抽样方法.

Combined classifier algorithm for imbalanced datasets

In order to improve the performance of the minority class, a combined classifier algorithm is presented based on data pre- processing. Firstly, Tomek links method is applied to preprocess a dataset, in which all Tomek links data points are removed to form a new dataset. Then the data points ofthe majority ctass in the new dataset are split into several disjoint subsets according to the imbalanced ratio, and each subset is combined with minority class to form a new training dataset. Finally, each training dataset is trained by least squares support vector machine(LS-SVM), and all of the LS-SVM classifiers are combined to form a classifying system. The label of a new testing data point is determined based on the voting strategy. The experimental results show that the proposed algorithm performs better than LS-SVM, synthetic minority over-sampling technique(SMOTE)and under-sampling(US)in terms of the classification performance of the majority class and the minority one.