Iterative Multi-mode Discretization: Applications to Co-clustering

Abstract

We introduce a new concept called “Iterative Multi-Mode Discretization (IMMD)” which is a new type of efficient data sparsification that can scale up many tasks in data mining. In this paper we demonstrate the application of IMMD in co-clustering, i.e. simultaneous clustering of the rows and columns in a matrix. We propose IMMD-CC, a novel co-clustering algorithm, which is developed based on IMMD. IMMD-CC has attractive properties. First, its time complexity is linear, so it can be used in large-scale problems. In addition, IMMD-CC is able to estimate the number of co-clusters automatically, and more accurate than state-of-the-art methods. We demonstrate the performance of IMMD-CC in comparison to several state-of-the-art methods on 100 data sets from a benchmark cohort, as well as 35 real-world datasets. The results show the promising potential of the proposed method.