Study of Continuous K-nearest Neighbor Algorithm
Download as PDF
DOI: 10.23977/cii2019.08
Corresponding Author
Yidong Song
ABSTRACT
In the machine learning algorithm, the K-nearest neighbor is mainly used to classify instances, and the selection of the value of K is especially critical. For the selection of the value of ķ, if the value is too large, the model is too simple to extract the law; if it is too small, the model is too complicated and over-fitting may occur. In general, we use cross-validation to get a good value of K. So that the model we built can better characterize the law of nature. This is only a fuzzy choice in the case of unclear data, and there is no guarantee that the model has strong practical value. Here, we introduce the "degree of structural law fluctuations", which can reflect the influence of different classifications on the overall law of things, and quantitatively express the objective laws of things for the first time, thus improving the accuracy and practical value of the K-nearest neighbor algorithm model. We call this algorithm as the improved algorithm, which is called continuous k-nearest neighbor algorithm.
KEYWORDS
K-nearest neighbor, continuous, prediction