Clustering Spatial Data: Addressing Spatial Autocorrelation and Multicollinearity Challenges

Clustering techniques for spatial data combine spatial and non-spatial attributes to form clusters of nearby spatial units with similar characteristics. When applied in regional economics, this concept can be adapted to optimize the formation of economic growth zones. It is expected that regions within one zone or cluster will exhibit similar economic activities and strong interactions, thereby accelerating economic growth. However, in this context, the non-spatial attributes are often intercorrelated, and it is not surprising that each attribute displays a spatial pattern. Consequently, using classical clustering methods may lead to suboptimal clusters. Therefore, the objective of this study is to modify the K-means clustering algorithm to accommodate spatial autocorrelation and multicollinearity. The clusters are formed based on the principal components of the local Moran’s index for each non-spatial attribute or variable. A simulation study is conducted to assess the performance of the modified technique. Generated spatial data, featuring various combinations of spatial autocorrelation and multicollinearity, based on East Java’s geographical conditions and regional economic performance, are utilized. The simulation study demonstrates that the modification performs well in producing clusters of contiguous regions.