Attributional Ruletrees: A New Representation for AQ Learning




Michalski, Ryszard S.

Journal Title

Journal ISSN

Volume Title



Attributional ruletrees are proposed as an extension of the current ruleset representation used by AQ type learning. The ruletrees split a multiclass classification problem into separate subproblems using a class splitting attribute. The resulting representation can be graphically represented as a tree whose root is assigned the class splitting attribute, branches stemming from the root are values (or sets of values) determining subsets of classes, and leaves are assigned ruleset families for classifying events to classes in these subsets. The values on the branches from the root thus define preconditions for applying ruleset families. Ruletrees are easy to interpret and understand, and can be generated by a relatively simple modification of the AQ algorithm presented below.




Michalski, R. S., "Attributional Ruletrees: A New Representation for AQ Learning," Reports of the Machine Learning and Inference Laboratory, MLI 02-1, George Mason University, Fairfax, VA, October, 2002 (slightly edited in May, 2004).