Practical Applications of Data Mining

96 Chapter 3 ClassifiCation learning

Notice that each of the above rules in the list has an accuracy of 100%.

They are potentially over-fitted. Next, we are going to post-prune the rules

as introduced with Induct. Rules 1, 5, and 6 have only one conditional term

and will not be considered in the pruning process. Rules 2, 3, and 4 have

three conditional terms each. We will use the probability measure as the

criterion to greedily trim off the over-fit conditions of these rules. Let’s start

with rule 2. As stated above, rules cannot be used independently. The real

meaning of rule 2 is as follows:

“If tear production rate is not reduced ...