How to find boundaries using clustering

Prior to clustering, you need to import a vector or raster data set. For point data, you should check the spatial network and edit the network if necessary. If you want to do clustering on classified data, create fuzzy classes from the original data set. Now you are ready to delineate clusters:

1. In the BoundarySeer window, go to the "Boundary" menu and choose "Detect Boundary," and then "Constrained Clustering."

2. The constrained clustering dialog consists of two tabs, "General" and "Advanced" settings.

3. "General" tab

   1. Select the data set that you wish to analyze, and select a name for the output boundary file.

   2. In the box marked "Number of clusters", enter an integer value for how many clusters you want the program to identify. You may wish to first perform a goodness of fit analysis to find the optimal cluster number for the data set. As the target cluster number sets the outcome, the choice is influential.

   3. Choose which variables to analyze from the data set. The default is to use all variables, and to give them all equal weights. If you want to use only one variable, you can fill in the dot next to "Variable" (rather than "Variable set") and select it. You may also select a subset of the variables and/or weight them.

   4. If you have more than one variable in your data set, you will have the option to standardize your data. If you plan to use the Steinhaus metric, you should not standardize.

4. "Advanced" tab

   1. Choose a dissimilarity metric from the pull-down list. For categorical data, the mismatch metric is the only option.

   2. Next, choose a method for agglomerative clustering.

      1. If you choose linkage clustering, choose which linkage method to use through setting the connectedness parameter. Connectedness values can range between 0 and 1, but they cannot equal 0 or 1.

      2. If you are doing linkage clustering with a large number of locations (e.g., a large raster data set) and want to subsample your clusters, enter your subsampling criteria.

   3. If you want to "cluster with k-means refinement," check the appropriate box.

5. Click "OK" at the bottom of the dialog. If you checked the standardization box, the standardization dialog box will appear. Here you should choose a standardization method, and decide where to store the modified data set.

6. BoundarySeer will ask if you wish to display the boundaries in the map. You can show the boundary in an existing map or create a new one. You can also view and manipulate the results as a table.