中文摘要：

KNN是基于实例的算法，对于大规模样本算法分类性能不高．针对这一缺点，提出一种基于概率模型的学习矢量量化神经网络的改进KNN分类新方法．考虑到最优参考点训练的重要性，结合概率方法得到最佳参考点的判断准则函数，采用梯度下降最优化算法利用LVQ训练参考点的最佳位置．在对未知样本进行分类时选出样本x的K个近邻，采用“投票选举”机制最后判断样本x的所属类别．新方法减少KNN的计算复杂度和时间，弥补了KNN在处理大规模数据问题上的不足．在UCI中数据集上的仿真实验表明改进算法的可行性．

英文摘要：

KNN is an algorithm based on living engineering practice, so that its classification performance is not high enough for large-scale sample. In view of this shortcoming, a novel pattern classification meth- od of improved KNN was presented on the basis of probability model-based learning vector quantized neu- ral network. Taking the importance of optimal reference point training into consideration, the probability method was combined to obtain the judgment criterion function for the optimal reference point, and the optimal position of reference point was trained with gradient-down optimization algorithm and LVQ. K- nearest neighbor of sample ：c was chosen for classification unknown sample and a ＂vote-election＂ mecha- nism was used to judge the class of the sample x. By using this new method, the complexity and time of KNN computation was decreased and the shortcoming of KNN in dealing with large-scale data was make up. It was shown by the simulation test on data set in UCI that the improved algorithm was feasible.