Requests

I'm looking for information on ruleset metrics. Clearly, in many situations, a classification ruleset of length n is preferred to one of length m for n < m (i.e. the model that builds 100 rules where another model with similar predictive capability builds 1000 rules is preferred).

Are there other metrics on rule sets besides this? Is there a specific term for this:

You have two rulesets with an equal number of rules and equal predictive capability. Each ruleset has 100 rules (for example). In the first ruleset, each rule applies to roughly 1% of the data. In the second ruleset, the first couple rules apply to 80% (another arbitrarily large number...). Is there a specific term for preferring the second ruleset?

(Editor Note: Please reply on KDnuggets Forums,

www.kdnuggets.com/phpBB/viewtopic.php?p=402

or to terry at letsche dot net