I have a ~ 100000 point shapefile. Now I need to take sample from this points while spatial auto correlation is least. I divided the point area by several square grid spacing like 100by 100 etc. and populated attribute with number of points inside each of these grids with intention of taking one point from each of this grids as a representative sample for logistic regression input. I checked the auto ccorrelation among the grid centers if it is negetive or at least zero. For this purpose I drew Moran I correllogram.But this process fails since I found 0 value at a distance which is too much and at that distance of grid spacing I will have only 5 grids so 5 points by which logistic tegression is impossible. So could anyone help me with best method for finding optimum grid dimension/spatial scale?
More background can be found at these links.
http://gis.stackexchange.com/questions/233799/how-to-properly-address-autocorrelation-for-logistic-regression-of-spatial-data
http://stats.stackexchange.com/questions/268335/how-to-take-sample-which-is-not-spatially-autocorrelated