RUSE-WARMR: Rule Selection for Classifier Induction in Multi-relational Data-Sets
Carlos Abreu Ferreira, João Manuel Portela da Gama, Vı́tor Santos Costa · 2008
One of the major challenges in knowledge discovery is how to extract meaningful and useful knowledge from the complex structured data that one finds in scientific and technological applications. One approach is to explore the logic relations in the database and using, say, an inductive logic programming (ILP) algorithm find descriptive and expressive patterns. These patterns can then be used as features to characterize the target concept. The effectiveness of these algorithms depends both upon the algorithm we use to generate the patterns and upon the classifier. Rule mining provides an excellent framework for efficiently mining the interesting patterns that are relevant. We propose a novel method to select discriminative patterns and evaluate the effectiveness of this method on a complex discovery application of practical interest.