Classification accuracy improvement using conditioned Latin Hypercube Sampling in Supervised Machine Learning

Abstract

The present study aims to elaborate on advanced sampling mechanisms employment into bleeding edge machine learning processes. Application of such approaches are considered in two stages. First, selection of a fit for purpose subset of the initially accumulated data to train the selected machine learning algorithm. Second, replacement of random sampling with samples drawn using Latin Hypercube Design (LHD) mechanisms. Latin Hypercube Sampling (LHS) was recently proposed as a sampling method based on covariates. The method provides full coverage of the vary of every variable by maximal stratification of the marginal distribution. Initial results on classification tasks are promising.

Topics & Concepts