| Abstrakt: |
As the variety and quality of spatial data increase in recent times, the potential to analyze local characteristics based on spatial data is getting stronger. Previous spatial analysis methods structuralize the spatial autocorrelation of data by the distances between data observation points and the contiguity of the data-observed regions. It is significant for the estimation of global characteristics of spatial data. However, these approaches are not suitable for identifying local differences from the data since they assume a smooth spatial autocorrelation structure. Generalized fused lasso, which can detect local differences in spatial data, has been proposed in machine learning studies. Its limitation is that the estimated parameters are biased toward zero; however, methods that overcome the limitation have also been proposed. Fused-MCP is one of those methods and is expected to be useful in spatial analyses. This study applies fused-MCP to spatial analyses. As an example of spatial analyses based on fused-MCP, this study analyzes the structure of geographical segmentation of the real estate market in central Tokyo. Fused-MCP is utilized to extract areas where the valuation standard is the same. The results reveal that the geographical segmentation displays hierarchal patterns. Specifically, the market is divided by municipalities, railway lines and stations, and neighborhoods. The case study confirmed the applicability of fused-MCP to spatial analyses. |
Full text from SpringerLink