一种基于最小最大邻域阶构图的半监督分类法

A Semi Supervised Classification Algorithm Based on Minimum and Maximum Neighborhood Order Composition

摘要: 为克服K近邻图边的对称问题及互K近邻图的连通性的不足，并且针对局部全局一致性学习(LLGC)算法的分类精度在很大程度上取决于控制参数α的设置，设置不合理可能造成分类的准确率较低，聚类的结果不准确的情况，研究提出一种半监督学习分类算法，将最小最大邻域阶构图法(KMM)结合少参数的简洁局部全局一致性学习(BB-LLGC)，即KMM-BB-LLGC算法，兼顾边的对称及整个图的连通，简化图上的目标函数，使其不受参数α的影响。用UCI数据库中的数据集进行实验，与KNN-LLGC、KNN-BB-LLGC、KMM-LLGC几种分类方法进行对比，实验表明，提出的方法能可以带来更高的分类准确率，达到较高的分类精度，算法效率更高，可以实现对样本精确、快速的分类。

Abstract: In order to overcome the problem of edge symmetry of the K-Nearest Neighbor Graph and the lack of connectivity of mutual K-Nearest Neighbor Graph, and the classification accuracy of local-global consistency learning (LLGC) algorithm largely depends on the setting of control parameters α, Unreasonable setting may result in low accuracy of a classification and inaccurate results of clustering. A semi-supervised learning classification algorithm is proposed, which combines the minimum and maximum neighborhood order composition method (KMM) with a kind of barebones LLGC (BB-LLGC) algorithm with fewer parameters, that is, KMM-BB-LLGC algorithm, considering the symmetry of the edge and the connectivity of the whole graph, simplifies the objective function on the graph and make it independent of parameters α, was used in experiments with data sets in UCI database. Compared with KNN-LLGC, KNN-BB-LLGC, KMM-LLGC, experiments show that the proposed method can bring higher clustering accuracy and achieve higher classification accuracy. It is more efficient and can realize the accurate and fast classification of samples.