An efficiency comparison of distance measures in K-Nearest Neighbors imputation for spatial data

Name: Prawit Banjong

Organization : Chulalongkorn University. Chulalongkorn Business School

Email : prawit.banjong@gmail.com

Name: Seksan Kiatsupaibul

Organization : Chulalongkorn University. Chulalongkorn Business School

Email : seksan@cbs.chula.ac.th

keyword: K-Nearest Neighbors Imputation

ThaSH: Management information systems

; Distance Measures

ThaSH: Spatial data infrastructures

Abstract: The spatial data was used in several areas. But the geography, topography and measuring machine limitation cause the missing data problem in any spatial data set and unable to analyze in association and causation. The objective of this study is to compare the efficiency of distance measures for the K-Nearest Neighbors (KNN) imputation method. The distance measures under this study include 1) Euclidean distance (ECD) 2) Manhattan distance (MHA) 3) Canberra distance (CBR) 4) Sorensen distance (SRS) 5) Euclidean distance of Cartesian coordinates (EOC) and 6) Arc Length by Harversine function (ALH). The comparisons are done across the different sizes of missing values and across the different values of the constant K (Number of nearest neighbor). The study is performed on simulated data sets whose distributions replicate that of an actual hourly air quality data set retrieved from Department of Pollution Control. The statistical hypothesis tests show that the efficiencies corresponding to the missing data imputation among distance measures and the numbers of nearest neighbors are significantly different. The results show that the distance measures can be divided into 3 groups. The first group consists of ECD, MHA, and CBR, the second group consists of SRS and the third group consists of EOC and ALH. The efficiencies between the three groups are significantly different, but the efficiencies within group are similar. In addition, the efficiency of the first group is highest, followed by the third and second group respectively. Furthermore, the results also show that the number of nearest neighbors causes the difference in the efficiency. It is found that the higher number of nearest neighbors, the higher efficiency until the optimal number of nearest neighbors that the efficiency will be lower. The interaction between the distance measures and the number of nearest neighbors also results in the difference in the efficiency.

King Mongkut's University of Technology North Bangkok. Central Library

Address: BANGKOK

Email: library@kmutnb.ac.th

Created: 2022

Modified: 2025-09-16

Issued: 2025-09-16

บทความ/Article

application/pdf

BibliograpyCitation : In King Mongkut's University of Technology North Bangkok Faculty of Applied Science, Thai Statistical Association (TSA) and Statistics Cooperative Research Network (Statistics CRN). The Proceeding of International Conference on Applied Statistics (ICAS 2022) (pp.173-178). Bangkok : King Mongkut's University of Technology North Bangkok

eng

©copyrights King Mongkut's University of Technology North Bangkok

ลำดับที่.	ชื่อแฟ้มข้อมูล	ขนาดแฟ้มข้อมูล	จำนวนเข้าถึง	วัน-เวลาเข้าถึงล่าสุด
1	ICAS 2022pp.173-178.pdf	503.46 KB

ใช้เวลา

0.01575 วินาที