KNN Algorithm Based On Gaussian Kernel And Its Applications

K-nearest neighbor classification algorithm is one of the top ten algorithms in machine learning.A distance function commonly used in traditional KNN algorith-m is Euclidean distance,the disadvantage of Euclidean distance is that different features have the same effect on classification by default.To solve this problem,this paper proposes a KNN classification algorithm based on Gaussian kernel.The distance function used in the proposed algorithm is based on the distance function of Gaussian kernel.This distance function gives weights to all the features of the sample,and its weight function is based on Gaussian kernel calculation.The distance obtained takes into account not only the relationship between the sample and the classification,but also the influence of the distribution of the training sample on the classification.At the same time,when we predicting the category of test samples,the algorithm proposed in this paper gives a similarity function based on affinity function on the basis of voting method.The similarity function gives smaller weights to the nearest neighbor samples far away from the test samples,thus the method can weaken their influence on the classification results.For the bandwidth parameterand parameterof the algorithm,this paper selects eight data sets from the UCI database to classify,and compares the classification accuracy to determine the bestand.In order to verify the good classification performance of the proposed algorith-m,two experiments are carried out in this paper.In the first experiment based on the bestand,we firstly use different distance functions for six data sets selected from eight data sets,and compare their classification accuracy.Secondly,the eight data sets are classified by three decision rules to compare their accuracy.Finally,in order to prove that the proposed classification algorithm is superior to other classification algorithms,this paper selects five data sets from eight data set-s to experiment,and compares theirand₁score.The experimental results show that the proposed algorithm has the best classification effect.In the second ex-periment we apply the classification algorithm proposed in this paper to the divorce data set.After the values of the two parameters are determined in the algorithm,it is compared with the traditional KNN algorithm and other classification algorithms.The results show that the algorithm proposed in this paper performs best on the divorce data set.Hence,the proposed KNN algorithm based on Gaussian kernel performs well in classification.