Classification Based On Nearest Neighbor Component Analysis And Its Application

K-nearest neighbor is one of the most common and simple and effective classification algorithms in the field of machine learning,but its disadvantages are also obvious.The distance measurement method based on the method is generally Euclidean distance,which does not consider the impact of different dimensions on the classification results,and the high dimension often cannot produce good classification results.K-nearest neighbor classification performance can be achieved by the measurement of supervised learning to use the label information to learn a new measure to improve.As for metric learning,nearest neighbor component(NCA)is a kind of distance measurement method based on positive definite matrix computing.This paper is to solve classification problems,the current common machine learning classification model such as logistic regression,decision tree,neural network,support vector machine(SVM),etc.They have their own advantages in the application,but there are also some limitations.Data classification method of this paper is mainly based on NCA approach,combined with K-nearest neighbor algorithm for dimension reduction as well as classification.This paper design the dimension reduction of space distance estimation method to promote the precision of the measurement matrix estimation,thus improves the classification accuracy of classifiers.The main research contents and innovation points are as follows:(1)Study of the classification model based on neighbor composition analysis.In principle,the close neighbor component analysis can train the appropriate distance measure based on matrix at the same time limit for data dimension reduction,equal to map the sample points to a new low dimensional space.But the choice of low dimensional space is very important to measure matrix of the training results,will also affect the final classification result.Based on the NCA,on the basis of composition analysis,according to the objective function of convergence situation determines the spatial dimension;According to the principle of matrix singular value decomposition to determine the direction of the space and bandwidth estimation method.In addition,in the original algorithm by using kernel smoothing estimates,considering the different data sets adaptability to different forms of kernel function is different.Try to change in the application of the different form of kernel function and finally chose the optimal form.(2)Access the accuracy of space estimation method by simulation experiment.Generate the different distribution of simulated data,and using this estimation method for model training,and to estimate the spatial dimension and direction comparing with the real value,the rationality of the estimation method was verified.(3)Apply this classification model in the medical disease detection.In the era of medical big data,machine learning has gradually been used in disease diagnosis and screening.Applied this model to the disease diagnosis,can better able to excavate the value of clinical data,to improve diagnosis efficiency provide powerful basis,clinical decision making,etc.This paper applies the data classification method to the image of female breast cancer in intelligent nuclear fetal health screening test,and the classification of the common classification model results were compared,having a certain practical significance to disease screening.