Is rule induction simply the process of applying any algorithm that discovers rules in data (e.g. neural networks, decision trees), or is it a technique in it's own right? If the latter can anyone explain one or two rule induction algorithms out there?
Rule Induction is separate to the rendering of rules from Decision Trees or Artificial Neural Networks and is an area within its own right.
Rule induction is effectively an optimisation problem given coverage and accuracy of the desired rules and that there is often a large potential rule set that would result.
A number of algorithms are available for providing solutions to such optimisation problems, including Genetic Algorithms, Simulated Annealing and Tabu Search - I would suggest a websearch for these.
In the sense that "rule" is used in "rule induction", neural networks generally do not find "rules". Here, "rule" means just what it does in everyday use: an IF..THEN statement about the data, like: IF substance = WATER AND temperature is ABOVE BOILING THEN state = GAS.
Most commercial logical learning systems are actually tree-induction systems (CART, See5, Scenario, etc.). True rule induction systems (such as WizWhy) are much less common and their resulting rule sets are not forced into the tree format. Trees cover all possible cases, whereas rules in general may not. Rules may also conflict (two or more rules may apply to a single case, and their THEN sides may not be the same), whereas this is not possible with trees.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.