基于LLRKNN算法的不平衡数据集分类应用

Application of Imbalanced Dataset Classification Based on LLRKNN Algorithm

不平衡数据集的特点是类样本数量差异比较大,K近邻(K-Nearest Neighbor,KNN)算法在对这种数据集分类时,容易出现多数类偏向,即容易将少数类识别为多数类。LLRKNN算法是为了降低多数类偏向的影响,对K近邻样本进行重构得出权值,算法分类决策由K近邻样本的权值决定。实验结果表明,LLRKNN算法对不平衡数据集的性能优于KNN算法,具有更好的稳定性。

Unbalanced data sets are characterized by large differences in the number of class samples. K-Nearest Neighbor(KNN) algorithm is prone to majority class bias when classifying such data sets, that is, it is easy to identify minority classes as majority classes. LLRKNN algorithm is designed to reduce the influence of most class bias. The weights of K-nearest neighbor samples are reconstructed. The classification decision of LLRKNN algorithm is determined by the weights of K-nearest neighbor samples. The experimental results show that the performance of LLRKNN algorithm for unbalanced data sets is better than that of KNN algorithm and has better stability.