Goodness of fit

Choose Detect Boundary > Constrained Clustering from the Data menu OR after right clicking on the dataset.

1. General Tab:

   1. Data: leukemia.

   2. Check the box next to "measure goodness of fit for multiple partitions"

   3. The clusters dataset and boundary names no longer apply for goodness of fit.

   4. Choose a minimum and a maximum number of clusters.

      • Minimum: 2.

      • Maximum: 250. Usually the best fit occurs when each point is in its own cluster (so cluster number = number of observations in the data set). That is not a classification. Choose a maximum below the theoretical maximum.

   5. Choose to detect boundaries using a single variable. Choose Cases/1000.

   6. Choose not to standardize data before detection (clear the checkbox).  Data standardization is only necessary when you analyze multiple variables.

2. Advanced Tab:

   1. Retain Squared Euclidean distance.

   2. Set the connectedness parameter to 0.8.   This setting asks BoundarySeer to compare adjacent clusters based on the more dissimilar elements in each cluster before deciding whether to merge them. This is a more stringent setting than lower connectedness.

3. Hit "OK".

4. Assess the Goodness of fit plot.