The Research On Semi-supervised Classification Algorithm Based On Two Different Composition Method

Driven by the rapid development of information technology, semi-supervised learning of this new form of machine learning in recent years to produce and continue to grow, rich data mining, statistical research, but also for other disciplines to better achieve development It provides more opportunities and possibilities. Today, the data collected through various channels that people can continue to increase, changing the way that these data have been widely spread, accompanied by the arrival of the era of the Internet +, in such circumstances, large amounts of data implied The use of information constantly being excavated to create new value. Graph-based semi-supervised learning because of its intuitive scholars have gradually been more studied and used. As a graph-based semi-supervised learning an important component part of appropriate composition determines the level of efficiency of the learning algorithm. Based on current research and learn the method of Graph-based Semi-supervised learning(GSL)mostly based on a patterning method--K neighbor patterning method, study other aspects of patterning method is relatively small. And because some of the commonly used patterning method to solve the existing problems in the sample data connection edges on the symmetry and connectivity. Therefore what kind of patterning method can better improve the research questions the efficiency of semi-supervised learning graph-based it is very urgent and necessary.This paper mainly from the following aspects:(1) review of a large number of domestic and foreign literature, combined with the current research and representative case studies, based on the FIG semi-supervised learning content and features to do a more comprehensive and complete review, and outlines the graph-based semi oversight basic theory and technology learning methods;(2)In order to overcome a semi-supervised classification algorithm commonly used in k neighbor figure composition method can’t meet the symmetry of the edge and figure of the lack of connectivity, this paper will be the minimum maximum neighborhood order method is applied to a semi-supervised classification algorithm in the process of the composition of a picture, is proposed based on the minimum maximum neighborhood order half a supervised classification algorithm(KMMLGC algorithm). Based on random sample data and the data on the UCI data sets of simulation experiments, show the effectiveness of KMMLGC algorithm(3)Natural nearest neighbor composition method is a kind of adaptive, non parameter method of nearest neighbor search sample. Method in this paper, the natural nearest neighbor composition is applied to a semi-supervised algorithms in the process of the composition of a picture, put forward a semi-supervised classification algorithm based on natural nearest neighbor(3 NLGC algorithm), a random sample sets and the simulation experiments on UCI data sets show the effectiveness of 3 NLGC algorithm.(4)at the same time for 3 NLGC algorithm and KMMLGC algorithm on UCI data sets has carried on the contrast experiment, the experiment shows that 3 n algorithm is more efficient.