Regresar

Detecting Dependency between Discrete Random Variables and Application

Abstract:

Finding the relationship between two random variables is important because it brings insights to design appropriate probabilistic models for a given dataset. The problem with current measures like Maximal Information Coefficient (MIC) is that can take a considerable amount of time to compute it, thus faster methods are desirable. In this paper, we propose two novel measures capable to detect linear and nonlinear dependency between random variables. The proposed measures are used to design new correlation functions, in order to compare them with the ones in literature. These novel dependency measures have been applied to supervised feature selection to show their usefulness.