Active relational rule learning in a constrained confidence rated boosting framework

Mehr zum Buch

Boosting is a robust technique designed to enhance prediction accuracy in systems that learn from examples. While extensively studied in propositional learning, its application in relational learning has received less attention. The author introduces an effective boosted ILP-based relational learning system alongside an embedded active feature selection technique, achieving a significant reduction in learning time—up to three orders of magnitude—compared to leading ILP systems. This improvement is realized without sacrificing interpretability or predictive accuracy. Unlike traditional feature selection methods in relational learning, this approach actively determines relevant feature subsets based on the ongoing learning process and does not require transforming examples into a propositional representation. This innovative method addresses the challenges of relational learning while maintaining high performance and clarity in the resulting hypotheses.