SQL/MX Data Mining Guide

Mining the Data
HP NonStop SQL/MX Data Mining Guide—523737-001
4-9
Checking the Model
Figure 4-4 shows the final decision tree for the example business opportunity.
Changing the Process
Classification trees are used to predict or explain responses to categorical dependent
variables. If you had not been able to develop a classification tree with homogeneous
branches with respect to Cust_Left, you could now do any of the following:
•
Redefine the statement of the business opportunity.
The data analysis process might indicate new directions that offer more interesting
results.
•
Redefine the goal.
The goal is equal to Y if the customer had a zero balance for a period of 3 months.
This definition might need adjustment.
•
Add or remove columns in the mining view.
Some columns that do not contribute to the goal can be removed. Also, the initial
analysis might give new insight into columns that could be added.
•
Change the definition of derived columns.
For example, the average balance for the period of 3 months might be a better
choice than a zero balance for 3 months.
•
Change the mappings on the encoded columns.
Each iteration of the data mining process gives new insight into the changes you might
make for the next iteration.
Checking the Model
After you develop a model, you can check the model against the mining data.
Figure 4-4. Final Decision Tree
Marital Status
Single
No Yes
0 3
Married
No Yes
2 1
Widow
No Yes
1 1
Divorced
No Yes
1 5
Male
No Yes
0 5
Female
No Yes
1 0
Chldrn=0
No Yes
0 3
Chldrn>0
No Yes
0 0
Prune the tree here because
the remaining branches do not
yield a pattern.