期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2019
卷号：97
期号：5
页码：1644-1654
出版社：Journal of Theoretical and Applied
摘要：This paper aims to examine some nonparametric classification methods based on the nearest neighbors rule including k nearest neighbors (KNN), distance weighted k nearest neighbors (DWKNN), local mean k nearest neighbors (LMKNN), and pseudo nearest neighbors (PNN). In order to know the performance of each method, we apply it in case of credit scoring in Indonesia, especially related to a micro credit. For each method studied, we use the same parameter, i.e. the Euclidean distance. After evaluating some odd value k, it is known that each method achieves the optimum classification performance at different k values. KNN achieved the best performance value at k = 11 with total accuracy of 84.91%, while DWKNN achieved best performance at k = 15 which only reached 77.36%. LMKNN works well on k = 9 with an accuracy value of 84.91% and PNN which is a combination between DWKNN and LMKNN only has an accuracy classification of 83.02%. In the case of micro credit in Indonesia with samples from a government bank in Wonogiri district, LMKNN is able to perform better than other methods. With k = 9, the classification performance of LMKNN is the same with the KNN that is obtained at k = 11. Therefore by using LMKNN will reduce the time in determining the label class of a prospective borrower.
关键词：Nonparametric Classification; K Nearest Neighbors; Distance Weighted K Nearest Neighbors; Local Mean K Nearest Neighbors; And Pseudo Nearest Neighbors
