K⁃近邻算法（ K⁃nearest neighbor， KNN）是一种思路简单、易于掌握、分类效果显著的算法。决定K⁃近邻算法分类效果关键因素之一就是距离的度量，欧氏距离经常作为K⁃近邻算法中度量函数，欧式距离将样本的不同特征量赋予相同的权重，但是不同特征量对分类结果准确性影响是不同的。采用更能体现特征量之间相对关系的卡方距离度量作为KNN算法的度量函数，并且采用灵敏度法进行特征权重计算，克服欧氏距离的不足。分类实验结果显示，基于卡方距离的改进算法的各项评价指标优于传统的KNN算法。%The K⁃nearest neighbor ( KNN) algorithm is one of the classification algorithms, and it is obvious in clas⁃sification effect, is simple and can be grasped easily. One of the most significant factors which determine the effect of the K⁃nearest neighbor classification is distance measure. Euclidean distance is usually considered as the measure function of the K⁃nearest neighbor algorithm, it assigns the same weight to different characteristics of the samples, but different characteristic parameter has different influence on the accuracy of the classification results. This paper adopts Chi⁃square distance which can more reflect the relative relationship between characteristics as measure func⁃tion of KNN algorithm and uses sensitivity method to compute the characteristic weight, so as to overcome the short⁃coming of Euclidean distance. The result of classification test shows evaluation indexes of the improved algorithm based on Chi⁃square distance are better than the traditional KNN algorithm.
展开▼
