一种改进的K近邻法在模式识别中的应用

针对传统K近邻法的缺陷,改进的K近邻法首先对训练样本进行聚类,将样本的特征空间划分成若干满足一定条件的小超球体,然后依据最近间隔值在这些小超球体内搜索待分类样本的K个近邻点.算法通过特征选取,选出反映样本模式重要信息的特征,从而确保了聚类的质量.同时K近邻算法中引入的最近间隔值,既确定了近邻点的搜索半径,又保障了搜索的准确性.通过实验证实,该方法不但节省时间,还有较高的识别率.

Application of an Improved K Nearest Neighbor Approach in Pattern Recognition

In order to overcome shortcomings of traditional K nearest neighbor approach,firstly the training samples are congregated by cluster algorithm,so feature space of sample is divided into many spheres,then search for nearest neighbor in these spheres according to closest interval.Features that can represent important information of sample pattern are chosen by feature selection of algorithm,so it ensures the quality of clustering.At the same time,closest interval in K nearest neighbor algorithm decides search radius of nearest neighbor point and guarantees accuracy of search.Experiments prove that the method not only saves time,but also has accurate result of classification.