[1J3-J-2] Machine learning: bayesian models

[1J3-J-2-04] Sparse Bayesian Learning for Itemset Data

〇Ryoichiro Yafune¹, Hiroto Saigo¹ (1. Kyushu University)

Keywords:Sparse Bayesian Learning, Itemset Mining, Bayesian Rejection

Sparse bayesian learning can learn sparse solution for linear classification / regression problem. Although it has a number of advantages over non-bayesian approach, extension of it to non-linear model is non-trivial. In this paper, we employ itemset mining, and consider building sparse bayesian model on the binary occurrence matrix of items. We propose an iterative algorithm that can efficiently extract non-linear features while avoiding the entire enumeration. In computational experiments based on simulated dataset, our approach could correctly identify non-linearity in the dataset. In experiments using HIV dataset, we demonstrate the effectiveness of bayesian approach by rejecting samples with large estimated variance.