Grafting for Combinatorial Boolean Model using Frequent Itemset Mining

Œis paper introduces the combinatorial Booleanmodel (CBM), which is defined as the class of linear combinations of conjunctions of Boolean aŠributes. Œis paper addresses the issue of learning CBM from labeled data. CBM is of high knowledge interoperability but naı̈ve learning of it requires exponentially large computation time with respect to data dimension and sample size. To overcome this computational difficulty, we propose an algorithm GRAB (GRA‰ing for Boolean datasets), which efficiently learns CBM within the L1-regularized lossminimization framework. Œe key idea ofGRAB is to reduce the loss minimization problem to the weighted frequent itemset mining, in which frequent paŠerns are efficiently computable. We employ benchmark datasets to empirically demonstrate that GRAB is effective in terms of computational efficiency, prediction accuracy and knowledge discovery.