A Data Structure to Represent Association Rules based Classifiers

We tackle the problem of representing association rules for a prediction purpose. We approach this problem by introducing a novel data structure for representing association rules (now seen as classification/regression rules). Unseen cases are fitted into a graph like structure that avoids any type of sorting procedure. The graph indexes the items present in the rules so that only rules with the antecedent covered by the new case are visited. A detailed description of the data structures to store the association rules is given along with the most important steps of the algorithm. Benchmarking and discussion on the main features is also presented.