一种自适应k-最近邻算法的研究

针对传统k-最近邻算法(k-Nearest Neighbor, kNN)存在搜索慢的缺陷，提出了一种改进型的自适应k-最近邻算法。该方法在以测试样本点为中心的超球内进行搜索，对超球半径的生长进行采样，建立半径生长的BP神经网络模型，逼近半径变化函数，并用该函数指导超球体的生长。该方法有效地缩小了搜索范围，减少了超球体半径生长的试探次数，对处理稀疏数据集有明显的优越性。

Research on an Adaptive k-Nearest Neighbor Algorithm

An improved adaptive k-nearest neighbor algorithm is brought forward because the traditional k-nearest neighbor algorithm has certain limitation that its searching speed is slow. The approach searches a super ball for the k-nearest neighbors, which takes the testing sample as its center. According to the radius growth of the super ball and the numbers of samples in the super ball,a BP model will be built to approximate the changing function of the radius.Then the BP model is used to guide the radius growth. The approach can effectively reduce the searching range and decrease the time of the super ball growth, which is very fit for sparse datum set.