Generating Rules from Examples

T h i i work describes tools for generating decision trees that are opt i mized w i th respect to linearity and are more efficient than those generated by Bratko's A O C D L [3] The rule generator is specialized to as to obey stated constraints corresponding to the above two properties. Rule induction takes advantage of one of the expert's most reliable and highly developed skills [9 ] , teaching by example, and avoids the need to resort to dialogue-acquisition of rules, tradit ionally recognized as the bottle-neck problem of knowledge engineering. However, decision trees derived from situation-action pairs are inherently less descriptive for expressing concepts than first-order or multivalued logic used in other projects [16] [8] [5], The lack of descriptive power is primarily associated wi th the absence of quantified variables.