This paper presents a new geographically weighted regression analysis tool, based upon a modified version of a General Regression Neural Network (GRNN). The new Geographic General Regression Neural Network (GGRNN) tool allows for local variations in the regression analysis. The algorithm of the GRNN has been extended to allow for both globally independent variables and local variables, restricted to a given spatial kernel. This mimics the results of Geographically Weighted Regression (GWR) analysis in a given geographical space. The GGRNN tool allows the user to load geographic data from the Shapefile into the underlying neural networks data structure. The spatial kernel can be either a fixed radius or adaptive, by using a given number of neighboring regions. The Holdout Method has been used to compare the fitness of a given model. An application of the tool has been presented using the benchmark working-age deaths in the Tokyo metropolitan area, Japan. Standardized residual maps produced by the GGRNN tool have been compared with those produced by the GWR4 tool for validation. The tool has been developed in the .Net C# programming language using the DotSpatial open source library. The tool is valuable because it allows the user to investigate the influence of spatially non-stationary processes in the regression analysis. The tool can also be used for prediction or interpolation purposes for a range of environmental, socioeconomic and public health applications.
[1]
N. Currit,et al.
Inductive regression: overcoming OLS limitations with the general regression neural network
,
2002
.
[2]
Rajan Nataraajan,et al.
A neuro-computational intelligence analysis of the ecological footprint of nations
,
2009,
Comput. Stat. Data Anal..
[3]
A S Fotheringham,et al.
Geographically weighted Poisson regression for disease association mapping
,
2005,
Statistics in medicine.
[4]
Chris Brunsdon,et al.
Geographically Weighted Regression: The Analysis of Spatially Varying Relationships
,
2002
.
[5]
Holger R. Maier,et al.
Forecasting chlorine residuals in a water distribution system using a general regression neural network
,
2006,
Math. Comput. Model..
[6]
Luca Brocca,et al.
Coupling soil moisture and precipitation observations for predicting hourly runoff at small catchment scale
,
2014
.
[7]
Donald F. Specht,et al.
A general regression neural network
,
1991,
IEEE Trans. Neural Networks.