Applying SKCV-algorithm to chemical data sets

Abstract

SKCV-algorithm can be used in research areas other than normal geographical environment. In this paper we will examine SKCV-algorithms functionality in chemical context by using SLOO-algorithm with 4 different chemical data set. Testing is split into 2 different cases where the first case tests how much removal of the closest points affects the prediction performance. The second case will study if SKCV-algorithm can be used as prediction estimator for sampling procedure and was tested with RPSD-algorithm in chemical space. The removal of points in data sets with uneven distribution caused prediction performance to become unstable when the algorithm removed too many points. The results of testing done with chemical data sets did not fully answer question about sampling estimator but with small changes to data sets and algorithms, the results became somewhat similar to results in the original research paper.