面向不平衡数据集的一种改进的k-近邻分类器

心脏心律失常数据集的心电图(ECG)数据往往存在各心律失常类型下样本数量不平衡问题.针对此问题，提出了一种新的模式识别分类方法，即改进的基于核的差重建的加权k-近邻分类器(modified kernel difference-weighted k-nearest neighbor classifier， MKDF-WKNN)，通过引入修正因子对含样本数较多的类别进行权值抑制，对含样本数较少的类别进行权值的加大，并使用UCI心脏心律失常数据集对ECG数据进行分类.实验结果表明，提出的算法和其他一些基于KNN的算法如KNN，DS-WKNN，DF-WKNN和KDF-WKNN相比，对于不平衡的心律失常数据集的分类有更好的效果.

A Modified KNN Classifier for Unbalanced Dataset

The existing arrhythmia datasets are suffering from the unbalanced number of training sample for electrocardiogram(ECG) data due to the obvious difference among the sample number of different types. A novel KNN-based classification algorithm， i.e.， a modified kernel difference-weighted KNN classifier(MKDF-WKNN) was proposed， by introducing a correction factor to restrain the weights of the categories with more samples and increase the weights of the categories with fewer samples. The experiment was carried on the UCI arrhythmia dataset to classify the ECG data. The results show that， for unbalanced datasets the proposed algorithm is better than some other KNN-based algorithms such as KNN， DS-WKNN， DF-WKNN and KDF-WKNN， in terms of classification accuracy.