Regression Analysis in Machine Learning using conditioned Latin Hypercube Sampling

The main objective of this study is to provide details on advanced sampling techniques employed in state-of-the-art machine learning processes. The utilization of these techniques is focused in two phases. The first stage involves selecting a suitable subset of the initial high-dimensional data to train the machine learning algorithm. The second stage entails replacing random sampling with Latin Hypercube Design (LHD) mechanisms. LHD represents a sampling method grounded in covariates, which ensures comprehensive coverage of the variable range by adeptly stratifying the marginal distribution. Recent experiments conducted on classification tasks using LHD have yielded encouraging initial results. This study adds to the expanding research on advanced sampling techniques, and highlights the potential of conditioned Latin Hypercube Design (cLHS) sampling as a powerful tool for improving machine learning performance.