| d) Group the values of your outcome field | f) Narrow down your attribute selection | |
The MinedTree object can use numeric, date/time and discrete attribute fields to classify the outcome fields when using tree induction. However, there are two situations when you will need to group together values of discrete attribute fields. The first instance is when the number of discrete values exceeds the practical limits of values that can be used to develop useful trees. The second reason for grouping is that it is recommended that you keep the number of discrete groups to 10 or so for best results from the induction algorithms, particularly if the number of records (rows) in your data source table is relatively small (i.e. below 2000).
For numeric and date/time attributes, grouping the numeric values into threshold ranges at this stage can also speed up the induction process.
You can define the grouping of values either manually or automatically. Auto grouping will combine values of attribute fields in accordance with their effect on the outcome field.
Manual grouping of attribute values is similar to the manual grouping of outcome values. It is invoked by using the right-mouse click on the highlighted attribute of interest and selecting Grouping. Manual grouping can be used to fully define the grouping of values, or it can be used to modify the automatically generated grouping.