In propositional learning, boosting has been a very popular technique for increasing the accuracy of classification learners. In firstorder learning, on the other hand, surprisingly little attention has been paid to boosting, perhaps due to the fact that simple forms of boosting lead to loss of comprehensibility and are too slow when used with standard ILP learners. In this paper, we show how both concerns can be addressed by using a recently proposed technique of constrained confidencerated boosting and a fast weak ILP learner.We give a detailed description of our algorithm and show on two standard benchmark problems that indeed such a weak learner can be boosted to perform comparably to state-of-the-art ILP systems while maintaining acceptable comprehensibility and obtaining short run-times.
|Translated title of the contribution||Relational Learning Using Constrained Confidence-Rated Boosting|
|Title of host publication||Unknown|
|Pages||51 - 64|
|Number of pages||13|
|Publication status||Published - Sep 2001|