Dynamic nearest neighbours for generating spatial weight matrix

Abstract

Spatial weight matrix is an important aspect in spatial analysis. Selecting different spatial weight matrix for the same analysis method will eventually generate different results. The commonly used scenarios to create spatial weight matrix are contiguity based and distance based. However, these scenarios have their own problems. Contiguity based scenario like Queen and Rook has disadvantages of forming unconnected neighbours especially for sparse region like islands. Meanwhile, distance based scenario needs specific input parameters, which often requires exhausted trials or expert judgement to specify the parameters. For distance based k-Nearest Neighbours, the result will be asymmetric weight matrix that cannot be used for two-way interaction analysis. To overcome these problems, we propose a Dynamic Nearest Neighbours (DNN) algorithm. It uses different types of distance, which are coordinate distance and attributed distance. In the evaluation, DNN algorithm outperforms other techniques of Rook, Queen, and Α-Nearest Neighbours since it can be applied to both contiguous and sparse regions and produce two-way relations.