An Information-Theoretic Measure of Dependency Among Variables in Large Datasets

Abstract

The maximal information coefficient (MIC), which measures the amount of dependence between two variables, is able to detect both linear and non-linear associations. However, computational cost grows rapidly as a function of the dataset size. In this paper, we develop a computationally efficient approximation to the MIC that replaces its dynamic programming step with a much simpler technique based on the uniform partitioning of data grid. A variety of experiments demonstrate the quality of our approximation.