I am wondering if someone could point me to the direction of support vector machines being used for longitude latitude. It seems logical that the possible complexity in SVM would be great for modelling variables dependent of location.
In this case, I am trying to model depression dependent on US zip codes. My idea is to convert these zip codes to longitude/latitude, and then use an SVM.
Has this approach been used before, would someone recommend a different approach?
Sitharam, et al. (2006). Spatial Variability of Rock Depth in Bangalore Using Geostatistical, Neural Network and Support Vector Machine Models.
The paper above compared NN's, SVM's, and ordinary kriging. The paper states
The coordinates(X, Y) of each data were prepared as input of the model, while reduced level of rock value was the output from this model.
which sounds like the data structure you are looking to model. Although SVM did not perform the best in this case, SVM did better than ordinary kriging, and it can certainly be used on spatial data (along with many other geostatistical techniques).
So to answer your question directly, yes I have seen it used for this type of data before. You could try SVM, but I would try other geostatistical approaches as well such as kriging, which is quite common.
Additionally, I'm not sure what program you plan to use to analyze your data, but this stack exchange link provides some implementation possibilities for R using the kernlab library. Apparently there is no direct SVM implementation in ArcGIS
https://gis.stackexchange.com/questions/63688/vector-machine-classification-in-arcgis
